fix: add detailed diagnostics and sync LFS tokenizer

fix: restore is_lora_valid to baseline.py

fix(space): lazy-load lora at runtime and keep dashboard responsive

fix(space): ship lora_adapter assets and enable train runtime deps

fix: stabilize deployment, action safety, and benchmark reproducibility

Finalize OpenEnv submission (multi-mode server, strict scoring, uv lock)

Add Colab training notebook and wire real inference logic into Streamlit

Initial commit for EcoGrid-OpenEnv hackathon