feat: Sci-RAG Engine — Llama Index RAG pipeline with Semantic Scholar, citation engine & AI access layer

[IvyX79]

Summary

Adds a complete Llama Index-powered RAG backend service (rag-engine/) for scientific document understanding, alongside frontend integration routes.

New rag-engine/ Service

• Llama Index — Hierarchical document parsing and retrieval (bounty requirement)
• Document Manager — PDF/DOCX/TXT/MD ingestion with deduplication
• Semantic Scholar + arXiv — Search and import external references
• Citation Engine — Every answer cites sources with confidence scores
• AI Access Layer — Token-based secure document interaction
• FastAPI server — 7 endpoints (query, upload, list, search, import, cite, health)
• Docker support — Dedicated Dockerfile, docker-compose integration
• 17/17 tests passing

Frontend Integration

• Routes queries through rag-engine with graceful Chroma fallback
• Routes uploads through rag-engine with fallback
• RAG_ENGINE_HOST env var configured

Key Differences from existing solution

Feature	Current	This PR
Llama Index	❌	✅ Full integration
Semantic Scholar	❌	✅ Search + import
Citation confidence	❌	✅ Scored per source
Document unification	❌	✅ Uploads + external refs
AI access layer	❌	✅ Token-based auth
Backend server	❌ Next.js only	✅ FastAPI service
Tests	91 lines	175 lines, 17 tests

Closes ISAAC-497

[IvyX79]

This PR addresses the Algora bounty [ISAAC-497] Implement an enhanced RAG Pipeline for Scientific/Research Workflows (https://app.algora.io/isaac/bounties/clq18zr98000ejs0gt0nv7gwu)

Adds:

• Complete Llama Index RAG backend service
• Semantic Scholar integration for academic paper discovery
• Citation engine with stable keys
• Document management (upload, chunk, index, search)
• AI access layer with content generation
• REST API endpoints

22 files changed, 2,289 lines added. Mergeable ✅