v2: Major RAG accuracy improvements — BGE-M3 embeddings, hybrid retrieval, reranking, HyDE (#1)

Update .env.example with OpenRouter config

Add OpenRouter support to ingest.py (garble check + merge LLM calls)

Add OpenRouter provider to sidebar with free model options

Integrate OpenRouter as AI provider (chain.py) with nvidia/nemotron-3-super-120b-a12b:free default

Fix Greek hallucination: rewrite system prompt + add output guardrails

Sync from GitHub