[FEAT] : add PDF Image Parsing Memory Optimizations (OOM Prevention)

[hrshjswniii]

🔗 Related Issue

---

Closes #487

---

---

📝 What does this PR do?

---

This PR resolves the memory-exhaustion (OOM) risks during PDF ingestion by optimizing how images are parsed and captioned:

• Page-by-Page Generator: Modified extract_pdf_images to be a generator that yields images lazily, page-by-page.
• On-the-fly Captioning: Caption generation is now executed during the main page chunking loop in chunk_document. Raw image byte strings are processed, transformed into text descriptions, and freed immediately.
• Memory Leak Safeguards: Wrapped local image variable cleanup in a finally block to guarantee raw binary data is garbage-collected page-by-page even if captioning throws an exception.
• Robust Ingestion Fallback: Added an exception handler inside chunk_document so that if the captioning model/API fails (e.g. rate limit, network timeout), the image isn't dropped. Instead, it falls back to a placeholder text chunk Image on page X., ensuring content presence is preserved.
• Unit Testing: Added test_pdf_image_captioning_on_the_fly to verify the generator loop behavior, error fallbacks, and byte-clearing correctness.

---

---

🗂️ Type of Change

---

• 🐛 Bug fix
• 🔧 Refactor / code cleanup
• 🧪 Tests

---

---

🧪 How was this tested?

---

• Tested the affected API endpoints manually
• Added / updated tests (Created test_pdf_image_captioning_on_the_fly in backend/tests/test_chunker.py)
• Ran full backend test suite (.venv/Scripts/pytest — all 120 tests passed)

---

---

⚠️ Anything to flag for reviewers?

---

• The old in-place captioning list crawler generate_captions_for_chunks in vision.py is kept for API compatibility, but is now safely bypassed during PDF chunking because the returned list of chunks from chunk_document contains pre-captioned image descriptions without raw byte payloads.

---

---

✅ 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