feat(rag): add extraction and indexing of PDF image captions

[nancysangani]

🔗 Related Issue

Closes #438

---

📝 What does this PR do?

Completes image caption extraction and indexing for the RAG pipeline across two files:

backend/app/rag/vision.py (rewrite of existing stub):

• Adds extract_captions_from_pdf() — uses PyMuPDF bounding-box proximity to find
  the nearest text block below (or above) each figure. Zero cost, fully offline.
• Fixes the broken OpenAI hook (openai.Image.create was an image-generation call,
  not a vision call) — replaced with gpt-4o-mini Chat Completions vision API,
  guarded behind VISION_PROVIDER=openai + OPENAI_API_KEY in settings.
• Restructures caption_image() resolution order:
  OpenAI (if configured) → OCR (pytesseract) → size-aware placeholder.
• generate_captions_for_chunks() now respects pre-extracted image_caption
  written by the ingestion pass and only falls back to OCR/placeholder for gaps.
• Raw image_bytes are always popped from chunks in the finally block —
  they are never serialised into ChromaDB.

backend/app/services/document_ingestion.py (surgical addition):

• Adds a proximity caption pass between chunk_document() and store_chunks().
• Calls extract_captions_from_pdf(), matches captions to image chunks by
  (page, figure_index) order, and writes image_caption + bbox before
  store_chunks() / generate_captions_for_chunks() runs.
• Wrapped in try/except — a caption failure never blocks ingestion.

---

🗂️ Type of Change

• ✨ New feature
• 🔧 Refactor / code cleanup

---

🧪 How was this tested?

• Ran the backend locally (uvicorn app.main:app --reload)
• Uploaded a PDF with labelled figures; confirmed ChromaDB stored chunks
  with is_image=True and non-empty image_caption matching adjacent text
• Verified images with no adjacent text receive the size-aware placeholder
• Confirmed decorative images below 1,000 px² are skipped
• Confirmed raw image_bytes are never present in stored chunk metadata
• Confirmed existing text and table chunks are unaffected (no regression)

---

✅ Self-Review Checklist

• My branch is based on dev, not main
• I have not added any secrets / API keys
• I have not modified main branch or any HuggingFace deployment config
• My code follows the existing style (no unnecessary formatting changes)
• I have updated relevant docs / comments if needed

[nancysangani]

Hi @param20h, Please review the PR when you get a chance. Thanks!

[nancysangani]

@param20h please review this PR, all checks have passed and resolved the conflicts. Thanks!

[github-actions]

🎉 Congratulations on getting your Pull Request merged! 🎉

Thank you for contributing to PDF-Assistant-RAG as part of GSSoC '26! 🚀

Keep up the great work! ✨