Deploy Option A: Transition HF Space to lightweight OpenEnv Server

Auto-skip wandb when WANDB_API_KEY is not set

Add reward tuning, improved prompt, eval harness, and serving Dockerfile

Fix: Update Unsloth installation and improve path handling in training script

Fix: Explicit HF token handling and final GRPO batch alignment

Fix P1/P2: Batch config, prompt synchronization, and evaluation parser consistency

Add scripts for hero case finding, hero details retrieval, and training log plotting

Initial clean deploy commit