fix: clamp scores to [0.01, 0.99] in API boundary, UI, inference, and tests

fix: isolate FAISS per domain, dynamic Auto Pilot step count

feat: add legal_research domain, multi-domain switching, and README overhaul

fix: mount Gradio at root, add Prometheus metrics

fix(server): CORS, sanitization, concurrency, parameter override, mount at /ui

feat: complete Agentic RAG Gym implementation