refactor: Inference and readme

rewrite reward system

fix llm based reward output

fix: reward system

fix: introduce SCORE_EPSILON and clamp scores in run_episode and main functions

fix: auto-start env server in inference.py

refactor: update default model and API endpoint, enhance error handling, and add close method for compatibility

fix: provide fallback API key and add safety check for empty observations in inference client

Sync root inference.py with fixed python/inference.py

Refactor project structure and remove unused files; update inference script for clarity and compliance with environment variable handling.

feat: add example environment variables, update README, and enhance inference script for better error handling

feat: add baseline scores JSON, inference script, and update Dockerfile for improved project structure