Add detailed README

[Copilot]

Repository had no documentation — single-line # Axiom heading only.

Added

• Overview — token-reduction pipeline rationale and the six TokenMiser stages
• Architecture diagram — ASCII art: Browser → Go Backend → ML Service → Supabase
• Request flow — annotated step-by-step walkthrough of a chat message through cache → classifier → summariser → tier router → prompt optimizer → model cascade
• Project structure — annotated directory tree for all three services
• Tech stack table — libraries, models, and infrastructure per layer
• Environment variables — per-service tables covering required and optional config with defaults
• Getting Started — copy-paste setup for ML Service, Backend, and Frontend
• API reference — all HTTP endpoints (backend + ML Service) with auth requirements and example request/response shapes
• ML pipeline deep-dive — internals of each stage: FAISS two-layer cache, 12-signal context classifier, rolling summariser, 7-signal tier router, prompt optimizer, and streaming CASCADE detection
• Token metrics — explains token_breakdown fields returned on every chat response
• Configuration reference — full table of all config knobs with defaults and descriptions