Build PDF/OCR extraction worker

[Xyloth]

Add worker-backed PDF text extraction, OCR fallback for scanned/image records, page counts, per-page text storage, and extraction warnings tied to the document vault.

[Xyloth]

Pressure test against a local private case folder: two scanned PDFs totaling 94 pages returned zero extractable text, and one image/chart-heavy DOCX returned no raw text. OCR should cover rendered PDF pages, image-only DOCX content, page thumbnails, searchable text, and confidence/error reporting without uploading sensitive records.