Benchmark: add LLM-as-a-Judge + BERTScore (hackathon 30% accuracy criterion)

Improve latency: parallel LLM calls, embedding cache, client reuse

Add .gitignore, dataset metadata, retrieval layer, and latest web/graphrag updates

Fix #8: Update web API compare route — 3-pipeline (LLM-Only + Basic RAG + GraphRAG) with side-by-side metrics

Add benchmark API route for live benchmark runs from dashboard

Add multi-provider compare API route + providers listing endpoint

Add providers listing API endpoint

Add Claude-powered compare API route with dual-pipeline orchestration