feat(usage): rename metrics → tokens_embedded/pages_embedded + export token cost to Prometheus

Billing product model finalized (Deck #281): bill pages externally, record
tokens internally. Rename the data-plane metric literals to match the now-
canonical contract (Deck #284) — the control plane's METRIC_EVENT_NAMES is
already renamed, so the old names would be unmapped and never sync to Stripe.

Rename (values unchanged):
- embeddings_queries → tokens_embedded (value = real token count, already
  emitted by this PR; the unit upstream providers bill on).
- pages_chunks → pages_embedded (value kept as len(chunk_texts) interim;
  TODO(#282): real normalized "pages indexed" count — real pages for paginated
  types, chars/tokens-per-page constant otherwise — is deferred to the
  instrumentation card, this only lands the name/contract).
- All literals, log strings, docstrings, comments, the migration comment, and
  tests renamed; grep confirms zero old strings remain.

Observability (new): export embedding token cost to Prometheus as
astrolabe_embedding_tokens_total{provider,operation} (operation = index|query)
so the billed cost unit is visible in Grafana, not just the per-tenant billing
DB. Dedicated counter (doesn't inflate the existing chunk/request metrics) and
always-on (independent of USAGE_METERING_ENABLED, so OSS/self-host gets it).
Wired on both the indexing batch embed and the search query embed (query inside
the per-request cache-miss branch, so reused embeddings aren't double-counted).

Note: the rename orphans any pre-existing embeddings_queries/pages_chunks rows
in tenant app DBs (CP no longer maps them) — acceptable; pipeline is inert with
throwaway dev/sandbox data.

Deck #284 (folded into PR #875).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/unit/providers/test_provider_usage.py

@@ -1,6 +1,6 @@
 """Token-usage surfacing: the Provider ABC estimate default + SimpleProvider.
 
-The usage-metering hooks (Deck #67) bill ``embeddings_queries`` by tokens. Real
+The usage-metering hooks (Deck #67) bill ``tokens_embedded`` by tokens. Real
 providers report exact counts from their API response; providers without a token
 field (Simple, and the ABC default) fall back to a char-based estimate so the
 billable value stays non-zero and monotone with input size.

tests/unit/search/test_bm25_hybrid.py

@@ -85,6 +85,7 @@ def patched_search(monkeypatch):
 
     settings = MagicMock()
     settings.get_collection_name.return_value = "test_collection"
+    settings.get_embedding_provider_family.return_value = "mistral"
     monkeypatch.setattr(
         "nextcloud_mcp_server.search.bm25_hybrid.get_settings", lambda: settings
     )

tests/unit/server/test_semantic_metering.py

@@ -1,6 +1,6 @@
 """Unit tests for the search-path usage-metering helper (Deck #67).
 
-``record_search_usage`` records the billable ``embeddings_queries`` event for a
+``record_search_usage`` records the billable ``tokens_embedded`` event for a
 semantic search. These pin the value mapping (query token count), the flag-off
 no-op, the doc_types metadata bounding, and the best-effort failure path —
 covering the server-tool metering wiring without standing up the full
@@ -38,7 +38,7 @@ async def test_records_query_token_count(store_spy):
 
     store_spy.record_usage_event.assert_awaited_once()
     kwargs = store_spy.record_usage_event.await_args.kwargs
-    assert kwargs["metric"] == "embeddings_queries"
+    assert kwargs["metric"] == "tokens_embedded"
     assert kwargs["value"] == 42
     assert kwargs["enabled"] is True
     assert kwargs["metadata"]["user_id"] == "alice"

tests/unit/test_embedding_metrics.py

@@ -12,7 +12,10 @@ from __future__ import annotations
 import pytest
 
 from nextcloud_mcp_server.config import Settings
-from nextcloud_mcp_server.observability.metrics import record_embedding
+from nextcloud_mcp_server.observability.metrics import (
+    record_embedding,
+    record_embedding_tokens,
+)
 
 pytestmark = pytest.mark.unit
 
@@ -120,3 +123,31 @@ class TestRecordEmbedding:
         assert metric_sample(
             "astrolabe_embedding_requests_total", {**labels, "status": "error"}
         ) == pytest.approx(1.0)
+
+
+class TestRecordEmbeddingTokens:
+    """astrolabe_embedding_tokens_total — token cost split by index/query."""
+
+    def test_index_increments_by_token_count(self, metric_sample):
+        labels = {"provider": "tok-prov", "operation": "index"}
+        before = metric_sample("astrolabe_embedding_tokens_total", labels)
+        record_embedding_tokens("tok-prov", "index", 4242)
+        assert metric_sample(
+            "astrolabe_embedding_tokens_total", labels
+        ) == pytest.approx(before + 4242)
+
+    def test_query_operation_is_separate_series(self, metric_sample):
+        labels = {"provider": "tok-prov", "operation": "query"}
+        before = metric_sample("astrolabe_embedding_tokens_total", labels)
+        record_embedding_tokens("tok-prov", "query", 7)
+        assert metric_sample(
+            "astrolabe_embedding_tokens_total", labels
+        ) == pytest.approx(before + 7)
+
+    def test_zero_or_negative_is_noop(self, metric_sample):
+        labels = {"provider": "tok-noop", "operation": "index"}
+        record_embedding_tokens("tok-noop", "index", 0)
+        record_embedding_tokens("tok-noop", "index", -3)
+        assert metric_sample(
+            "astrolabe_embedding_tokens_total", labels
+        ) == pytest.approx(0.0)

tests/unit/test_processor_metering.py

@@ -1,7 +1,7 @@
 """Unit tests for the indexing-path usage-metering helper (Deck #67).
 
-``record_indexing_usage`` records the two billable events (``pages_chunks`` +
-``embeddings_queries``) after a document's chunks are embedded. These cover the
+``record_indexing_usage`` records the two billable events (``pages_embedded`` +
+``tokens_embedded``) after a document's chunks are embedded. These cover the
 value mapping, the flag/zero-chunk no-ops, and the best-effort failure path
 without standing up the full document pipeline.
 """
@@ -25,8 +25,8 @@ def store_spy(monkeypatch):
 
 
 @pytest.mark.unit
-async def test_records_pages_chunks_and_token_count(store_spy):
-    """Both events fire: pages_chunks = chunk count, embeddings_queries = tokens."""
+async def test_records_pages_embedded_and_token_count(store_spy):
+    """Both events fire: pages_embedded = chunk count, tokens_embedded = tokens."""
     await processor.record_indexing_usage(
         enabled=True,
         provider="mistral",
@@ -40,7 +40,7 @@ async def test_records_pages_chunks_and_token_count(store_spy):
 
     calls = store_spy.record_usage_event.await_args_list
     by_metric = {c.kwargs["metric"]: c.kwargs["value"] for c in calls}
-    assert by_metric == {"pages_chunks": 110, "embeddings_queries": 4242}
+    assert by_metric == {"pages_embedded": 110, "tokens_embedded": 4242}
     for c in calls:
         # Hot-path fast-gate + tenant-local attribution metadata.
         assert c.kwargs["enabled"] is True

tests/unit/test_usage_store.py

@@ -95,7 +95,7 @@ async def test_flag_off_is_noop(storage, monkeypatch):
     """With metering disabled, nothing is written (zero DB work)."""
     _set_metering(monkeypatch, False)
     store = UsageEventStore(storage)
-    await store.record_usage_event(metric="pages_chunks", value=5)
+    await store.record_usage_event(metric="pages_embedded", value=5)
     assert await _count(storage) == 0
 
 
@@ -115,12 +115,12 @@ async def test_enabled_param_short_circuits_without_reading_settings(
     monkeypatch.setattr(store_module, "get_settings", _boom)
     store = UsageEventStore(storage)
 
-    await store.record_usage_event(metric="pages_chunks", value=1, enabled=False)
+    await store.record_usage_event(metric="pages_embedded", value=1, enabled=False)
     assert await _count(storage) == 0
 
     eid = str(uuid.uuid4())
     await store.record_usage_event(
-        metric="pages_chunks", value=1, event_id=eid, enabled=True
+        metric="pages_embedded", value=1, event_id=eid, enabled=True
     )
     assert await _count(storage) == 1
 
@@ -131,7 +131,7 @@ async def test_insert_roundtrip(storage, monkeypatch):
     store = UsageEventStore(storage)
     eid = str(uuid.uuid4())
     await store.record_usage_event(
-        metric="pages_chunks",
+        metric="pages_embedded",
         value=7,
         event_id=eid,
         metadata={"provider": "gateway"},
@@ -140,7 +140,7 @@ async def test_insert_roundtrip(storage, monkeypatch):
     assert row is not None
     # Postgres returns event_id as a uuid.UUID; normalize to str for compare.
     assert str(row[0]) == eid
-    assert row[2] == "pages_chunks"
+    assert row[2] == "pages_embedded"
     assert row[3] == 7
 
 
@@ -149,11 +149,11 @@ async def test_on_conflict_dedup(storage, monkeypatch):
     _set_metering(monkeypatch, True)
     store = UsageEventStore(storage)
     eid = str(uuid.uuid4())
-    await store.record_usage_event(metric="pages_chunks", value=1, event_id=eid)
-    await store.record_usage_event(metric="embeddings_queries", value=99, event_id=eid)
+    await store.record_usage_event(metric="pages_embedded", value=1, event_id=eid)
+    await store.record_usage_event(metric="tokens_embedded", value=99, event_id=eid)
     assert await _count(storage) == 1
     row = await _fetch(storage, eid)
-    assert row[2] == "pages_chunks"  # DO NOTHING, not DO UPDATE
+    assert row[2] == "pages_embedded"  # DO NOTHING, not DO UPDATE
     assert row[3] == 1
 
 
@@ -164,7 +164,7 @@ async def test_metadata_json_roundtrip(storage, monkeypatch):
     eid = str(uuid.uuid4())
     meta = {"provider": "gateway", "model": "titan", "nested": {"chunks": 3}}
     await store.record_usage_event(
-        metric="pages_chunks", value=3, event_id=eid, metadata=meta
+        metric="pages_embedded", value=3, event_id=eid, metadata=meta
     )
     row = await _fetch(storage, eid)
     raw = row[4]
@@ -187,7 +187,7 @@ async def test_occurred_at_roundtrip(storage, monkeypatch):
     eid = str(uuid.uuid4())
     when = datetime(2026, 1, 15, 12, 0, 0, tzinfo=timezone.utc)
     await store.record_usage_event(
-        metric="pages_chunks", value=1, event_id=eid, occurred_at=when
+        metric="pages_embedded", value=1, event_id=eid, occurred_at=when
     )
     row = await _fetch(storage, eid)
     stored = row[1]
@@ -207,7 +207,7 @@ async def test_metadata_none_is_null(storage, monkeypatch):
     store = UsageEventStore(storage)
     eid = str(uuid.uuid4())
     await store.record_usage_event(
-        metric="embeddings_queries", value=1, event_id=eid, metadata=None
+        metric="tokens_embedded", value=1, event_id=eid, metadata=None
     )
     row = await _fetch(storage, eid)
     assert row[4] is None
@@ -233,7 +233,7 @@ async def test_best_effort_swallows_db_errors(storage, monkeypatch, caplog):
 
     # Must not raise.
     with caplog.at_level(logging.WARNING, logger="nextcloud_mcp_server.usage.store"):
-        await store.record_usage_event(metric="pages_chunks", value=1)
+        await store.record_usage_event(metric="pages_embedded", value=1)
 
     assert recorded, "record_db_operation should be called on the error path"
     assert recorded[-1][3] == "error"
@@ -262,7 +262,7 @@ async def test_best_effort_swallows_unserializable_metadata(
     # Must not raise.
     with caplog.at_level(logging.WARNING, logger="nextcloud_mcp_server.usage.store"):
         await store.record_usage_event(
-            metric="pages_chunks", value=1, metadata=bad_metadata
+            metric="pages_embedded", value=1, metadata=bad_metadata
         )
 
     # Nothing was written — the encode failed before the insert.