Proposal: opt-in local OCR for image-only PDFs (#5)

[ducanhnguyen223]

Following up on issue #5, I would like to confirm the first-iteration boundaries before opening an implementation PR.

Proposed behavior:

• Keep pypdf/pdftotext as the default path with no new required dependencies.
• Offer explicit local OCRmyPDF/Tesseract opt-in; never send documents to a remote OCR service.
• Invoke OCR only when the PDF has no usable extracted text, keep page labels in source citations, and remove temporary sidecar files on success or failure.
• Keep searchable-PDF behavior unchanged and make normal tests independent of OCR binaries.

Should the first version intentionally leave mixed text/scanned PDFs to the existing extractor, or should it OCR only the scanned pages in a mixed PDF? Are there preferred optional-install or CLI conventions?

I have a local draft but will hold the PR until the scope and compatibility expectations are clear.