Switch PDF page rendering from external pdftoppm/pdftocairo subprocess calls
to github.com/gen2brain/go-fitz (MuPDF wrapper), eliminating the poppler-utils
runtime dependency. Enable CGO in Dockerfile builder stage and install gcc/musl-dev
for the static MuPDF link; runtime image remains unchanged.
- Downgrade go directive in go.mod from 1.23.0 back to 1.20 to match
Docker builder image (golang:1.20-alpine); re-run go mod tidy with
go1.20 (via gvm) to keep go.sum consistent
- GetPDFTask now verifies callerUserID matches task.UserID to prevent
cross-user data exposure of PDF page content
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Migrate recognition_results table to JSON schema (meta_data + content),
replacing flat latex/markdown/mathml/mml columns
- Add TaskTypePDF constant and update all formula read/write paths
- Add PDFRecognitionService using pdftoppm (Poppler) for CGO-free page
rendering; limits processing to first 10 pages (pre-hook)
- Reuse existing downstream OCR endpoint (cloud.texpixel.com) for each
page image; stores results as [{page_number, markdown}] JSON array
- Add Redis queue + distributed lock for PDF worker goroutine
- Add REST endpoints: POST /v1/pdf/recognition, GET /v1/pdf/recognition/:task_no
- Add .pdf to OSS upload file type whitelist
- Add migrations/pdf_recognition.sql for safe data migration
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
