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

Update __init__.py with new exports (TGGraphRAGClient, TripleComparisonResult, LLM judge, BERTScore, etc.)

Fix #3: Add LLM-as-a-Judge (PASS/FAIL) + BERTScore evaluation — the two hackathon-required accuracy metrics

Fix #2+#4+#6: Add LLM-Only pipeline, wire NoveltyEngine, integrate TG GraphRAG client, 3-pipeline comparison

Fix #1: Add TigerGraph GraphRAG integration layer wrapping official repo REST APIs

Add advanced GSQL queries: PPR scoring, shortest paths, spreading activation, community detection"

Add GraphRAG novelties engine: PPR scoring, token budget, path pruning, spreading activation, hybrid router, incremental updates

Wire UniversalLLM into layers __init__ + add benchmark API endpoint for live runs

Add Python universal LLM layer with LiteLLM supporting 12 providers + Ollama

Add Layer 2: Inference Orchestration (dual pipeline, adaptive routing, comparison)

Add Layer 4: Evaluation Layer (RAGAS + custom F1/EM metrics + benchmarking)

Add Layer 1: Graph Layer (TigerGraph schema, GSQL queries, vector search, traversal)

Add Layer 3: LLM Layer (generation, entity extraction, keyword extraction, complexity analysis)

Add layers init