fix(review): warn (not debug) on shadow-classify failure; tidy pymupdf usage

Address PR #855 round 2:

- 🔴 _shadow_classify swallowed all exceptions at DEBUG, so a systematic
  failure (pymupdf bug, memory pressure) is invisible at LOG_LEVEL=INFO and
  trips SonarQube S2221/S5754. Log at WARNING instead (still best-effort --
  indexing is unaffected).
- classifier: use `with pymupdf.open(...) as doc` instead of manual try/finally.
- tests: release the Pixmap's native memory (del pix) in the image fixtures.

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

nextcloud_mcp_server/document_processors/classifier.py

@@ -113,8 +113,7 @@ def classify_pdf(content: bytes) -> DocClassification:
     """
     import pymupdf  # noqa: PLC0415 -- keep the heavy import lazy / off module load
 
-    doc = pymupdf.open("pdf", content)
-    try:
+    with pymupdf.open("pdf", content) as doc:
         page_count = doc.page_count
         indices = _sample_indices(page_count)
         pages: list[PageSignals] = []
@@ -141,8 +140,6 @@ def classify_pdf(content: bytes) -> DocClassification:
             pages.append(
                 PageSignals(n, len(text), round(coverage, 3), quality, needs_ocr)
             )
-    finally:
-        doc.close()
 
     sampled = len(pages)
     total_chars = sum(p.char_count for p in pages)