extract_pdf_text returns "" silently on extraction failure (indistinguishable from an empty PDF) #4527

extract_pdf_text wraps its entire PdfReader loop in except Exception: return "", so a malformed, encrypted, or otherwise unreadable PDF is indistinguishable from a PDF with no text. Downstream consumers (knowledge-graph builders, RAG indexers) count the document as "empty" rather than "failed" — in our pipeline this was a coverage-audit blind spot: files producing zero nodes with no trace anywhere, undetectable without re-running extraction manually.

Suggested shape: a warning to stderr naming the file and the exception class before returning (or a distinct sentinel), and a separate warning when the result is near-empty (so scanned/image-only PDFs are visible in logs rather than silently skipped — related: the OCR gap in #181).

We implemented both locally (a warning on failure; a near-empty-result warning; an optional OCR fallback for the latter case) and they turned silent data loss into visible, countable events. Small patch, happy to PR.

Environment: graphifyy 0.9.92, Python 3.13, pypdf (pdf extra).