refactor(inference, train): unify system prompts and enhance user prompt structure

refactor(inference): enhance circuit breaker logic and model initialization

feat(docs): enhance README and PITCH with detailed Viveka overview

feat(eval): add AQI evaluation scripts and reliability diagrams

feat(inference, train): add baseline policies and training script for Viveka

feat(tests): update test_graders.py for Viveka's Phase 2 components

feat(ui): per-tier breakdown in full-bench comparison

feat(ui): full-bench comparison — naive vs heuristic across all scenarios

feat(ui): side-by-side policy comparison panel + 3 tests

feat(dependencies): add dev dependency group and update package markers

test(e2e): full-stack episode walkthroughs for 3 critical paths

feat(ui+scenarios): live-trace Gradio UI + 5 T2 medium scenarios

feat(env): scenario validator + expected_state matcher + 4 T1 scenarios

feat(services): full NPCI/DigiLocker/IRCTC semantics with 49 unit tests

scaffold viveka-env: 3-service DPI substrate (UPI + DigiLocker + IRCTC)