fix: reduce grpo training runtime

fix: parse chat completions in rewards

fix: emit prompt column for grpo trainer

fix: add grpo warnings compatibility attr

fix: choose writable cache root at space startup

fix: use writable runtime dirs in training space

fix: patch torchao dtype imports for unsloth

fix: shim torch._pytree.register_constant for torchao compat

fix: verbose per-step logs and BaseException catch in _run_training

fix: capture transformers/TRL/unsloth logs in Gradio UI

fix: install unsloth in Docker image at build time

fix: chunk-read pip output splitting on \r and \n

feat: stream pip install output live to UI logs

fix: install unsloth to /app/pkgs to bypass HF Space permission issue

fix: robust unsloth install — PyPI first, git fallback, log errors

feat: add A100 training Space with manual-trigger Gradio UI

feat: bulletproof GRPO training script + Colab notebook

feat: baseline eval and GRPO training script

feat: initial structure