Task 13: GRPO training notebook (Unsloth + TRL, 17 cells, base-vs-trained eval, plot cells, push-to-Hub)

Phase markers + restraint scoping: silent_witness/implicit_preference reward silence; secret_handling/multi_party expect engagement

Fix HF Space metadata: colorFrom yellow, shorter description

Merge HF Space init (keep our README + frontmatter)

initial commit

Initial KinChat env: models, personas, scenarios, rubrics, grader, env loop, FastAPI app, client, dashboard, baseline inference (377 tests passing)