docs(document-processors): round-2 review nits + classify_pdf glyph test

Address round-2 review on PR #914:
- Add corrupt_glyphs to the document_classifier_flag_total label comment (it is
  a live flag value emitted by record_document_classification).
- Mirror the full_text-vs-sampled control-ratio NOTE into classify_pdf so the
  diagnostic path's under-detection trade-off is documented in place.
- Add test_classify_pdf_glyph_corrupt_routes_structured for routing symmetry on
  the standalone classify_pdf path.

(SonarCloud quality gate is green — the prior S1244 finding was fixed last round.)

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

nextcloud_mcp_server/document_processors/classifier.py

@@ -272,6 +272,10 @@ def classify_pdf(content: bytes) -> DocClassification:
     ocr_frac = (sum(p.needs_ocr for p in pages) / sampled) if sampled else 0.0
     # Char-weighted doc-level control-char ratio (p.control_ratio * char_count is
     # the per-page bad-char count). The glyph-leak signal -- see _control_char_ratio.
+    # NOTE: this is over the <=MAX_SAMPLED_PAGES sample, so unlike classify_from_text
+    # (which scans the whole full_text) this diagnostic path can under-detect
+    # corruption concentrated outside the sampled pages. Acceptable here: the hot
+    # path is classify_from_text; this standalone pass is for diagnostics.
     control_ratio = (
         sum(p.control_ratio * p.char_count for p in pages) / total_chars
         if total_chars

nextcloud_mcp_server/observability/metrics.py

@@ -334,7 +334,7 @@ document_classifier_flag_total = Counter(
     # so flag{image_heavy} is expected to exceed classified{recommended_tier=ocr}.
     "astrolabe_document_classifier_flag_total",
     "Tier-0 classifier flags raised on documents",
-    ["flag"],  # image_heavy | scanned | bad_text_layer
+    ["flag"],  # image_heavy | scanned | bad_text_layer | corrupt_glyphs
 )
 
 document_text_quality = Histogram(