Update arbitragent Colab GRPO config

Colab: GRPOConfig bf16/fp16=False, try/except for use_bf16

fix colab repo url

negotiation bluff classifier + message cleaner

Add test selfplay states for HF Spaces

Add test selfplay states for HF Spaces

Clean repo — code only, no weights or training data

Add all code, exclude large model weights

Add arbitragent_colab.ipynb — end-to-end Colab notebook for hackathon

Initial commit: ArbitrAgent with README, agent loop, envs, demo, training