Clean up a camera photo and return enhanced JPEG bytes

Run the document-digitalization pipeline (perspective correction + CLAHE

• unsharp mask) on a camera photo and return the result as a JPEG so the frontend can show it in the preview tile and feed it into the QR reader.

Stateless — no DB writes, no S3 writes. PDFs are rejected (the save-side flow already handles them). On any CV failure or unsupported input we return the original bytes so the frontend can stay simple — it always gets some bytes back.