chore: add PyMuPDF for PDF → image rendering

feat: PDF → images via PyMuPDF for Nemotron; both models PDF-only

fix: syntax error in dragleave handler breaking file upload

chore: add openai SDK to requirements

feat: dual model support — Gemma 4 + NVIDIA Nemotron 3 with OpenAI SDK, multimodal, retry-fallback

Update README.md

revert: back to pypdf from docling

relax: remove docling version constraint

feat: docling multi-format parsing with clean history