Add multipart/form-data and signed URL support to /glmocr/parse

[majcheradam]

Summary

• server.py: /glmocr/parse now accepts multipart/form-data (uploaded files via files field, remote URLs via urls field) and http/https signed URLs in JSON mode; all inputs are converted to data: URIs in-memory — no temp files written
• image_utils.py: added pdf_bytes_to_images_pil and pdf_bytes_to_images_pil_iter for zero-disk-I/O PDF rendering via pypdfium2 in-memory bytes API
• page_loader.py: added data:application/pdf;base64,... routing in _load_source / _iter_source; added _load_pdf_bytes and _iter_pdf_bytes methods

Usage

File upload (multipart):

curl -X POST http://localhost:5002/glmocr/parse \
  -F "files=@document.pdf" \
  -F "files=@image.png"

Signed URL (JSON):

curl -X POST http://localhost:5002/glmocr/parse \
  -H "Content-Type: application/json" \
  -d '{"images": ["https://s3.amazonaws.com/bucket/doc.pdf?X-Amz-Signature=..."]}'

Mixed (multipart + remote URLs):

curl -X POST http://localhost:5002/glmocr/parse \
  -F "files=@local.pdf" \
  -F "urls=https://signed-url/remote.png"

Test plan

• Upload a PDF via multipart — verify all pages rendered correctly
• Upload an image via multipart — verify result matches JSON path
• Pass a signed S3/GCS URL in JSON images — verify fetched in-memory
• Pass a signed URL in multipart urls field — verify resolved correctly
• Existing application/json with file:// and data: URLs still works
• Invalid Content-Type returns 415

🤖 Generated with Claude Code