updated EICC v2 environment, APIs, and training pipeline

Run baseline on easy/medium/hard; fix /inference score from last [END]

Enhance customer support environment with reward breakdown and new dependencies. Added detailed reward breakdown for actions, including classification, routing, response, escalation, and resolution. Updated `pyproject.toml` to include new dependencies: OpenAI, FastAPI, Uvicorn, and OpenEnv core. Improved internal state management to track last reward breakdown for interpretability.