fix(review): unify "scanned" flag name + log image_coverage length drift

Address PR #863 round 3:

- classify_from_text emits the "scanned" flag (was "no_text_layer") for the
  empty-text-layer case -- same name + meaning as classify_pdf, so
  astrolabe_document_classifier_flag_total isn't split across two labels for the
  same concept (and matches the metric's documented vocab).
- classify_from_text logs at DEBUG when image_coverage length != the expected
  min(pages, MAX_SAMPLED_PAGES), so a 1:1-alignment contract break (extractor
  reorders/skips pages) surfaces instead of silently misattributing coverage.

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

nextcloud_mcp_server/document_processors/classifier.py

@@ -247,6 +247,22 @@ def classify_from_text(
     ``page_boundaries`` are ``{page, start_offset, end_offset}`` indexing into
     ``full_text``; ``image_coverage[i]`` (if given) aligns with the i-th boundary.
     """
+    # image_coverage is expected to be one entry per page, capped at
+    # MAX_SAMPLED_PAGES (see image_coverage_per_page). Any other length means the
+    # 1:1 page alignment drifted (e.g. the extractor reordered/skipped pages) --
+    # log it so a contract break surfaces rather than silently misattributing
+    # coverage to the wrong pages.
+    if image_coverage is not None:
+        expected = min(len(page_boundaries), MAX_SAMPLED_PAGES)
+        if len(image_coverage) != expected:
+            logger.debug(
+                "image_coverage length %s != expected %s for %s boundaries; "
+                "scan signal may be misaligned",
+                len(image_coverage),
+                expected,
+                len(page_boundaries),
+            )
+
     pages: list[PageSignals] = []
     for idx, b in enumerate(page_boundaries):
         seg = full_text[b["start_offset"] : b["end_offset"]]
@@ -283,7 +299,10 @@ def classify_from_text(
     flags: set[str] = set()
     if sampled and ocr_frac >= page_fraction:
         if total_chars == 0:
-            flags.add("no_text_layer")
+            # "scanned" (not "no_text_layer"): same name + meaning as classify_pdf
+            # so astrolabe_document_classifier_flag_total isn't split across two
+            # labels for the empty-text-layer case.
+            flags.add("scanned")
         elif mean_quality < min_text_quality:
             flags.add("bad_text_layer")
     if any(p.image_coverage >= IMAGE_COVERAGE_SCANNED for p in pages):

tests/unit/test_doc_classifier.py

@@ -174,7 +174,7 @@ def test_classify_from_text_clean_routes_fast():
 def test_classify_from_text_empty_routes_ocr():
     c = clf.classify_from_text("", [{"page": 1, "start_offset": 0, "end_offset": 0}])
     assert c.recommended_tier == "ocr"
-    assert "no_text_layer" in c.flags
+    assert "scanned" in c.flags  # unified with classify_pdf's flag name
     assert c.total_chars == 0
 
 
@@ -202,7 +202,7 @@ def test_classify_from_text_junk_layer_flags_bad_text_layer():
     assert c.recommended_tier == "ocr"
     assert c.total_chars > 0
     assert "bad_text_layer" in c.flags
-    assert "no_text_layer" not in c.flags
+    assert "scanned" not in c.flags  # has text, just junk -> not the empty case
 
 
 # --- quality + scan escalation triggers (Deck #207) --------------------------