feat: add support for lowercase Hugging Face Space secrets

Submission-ready: README, blog, training pipeline, baseline evidence, OpenEnv compliance

fix(grpo): Unsloth inference-mode swap + smaller LoRA + KV cleanup (T4 OOM #2)

fix(grpo): per-step backward to bound VRAM during update (T4 OOM fix)

feat(kaggle): add clean_launch.py + shrink budget to 20/25/30 = 75 eps

feat(kaggle): default to fixed-budget curriculum 20/30/50 episodes

fix(grpo): skip reference model when kl_beta=0 to save 5GB VRAM on T4

fix(kaggle): align pip-managed numpy with kernel's loaded numpy

fix(kaggle): pin torch via constraints file in REPAIR cell

fix(kaggle): escape backslash-n in REPAIR cell separator print

chore(kaggle): rebuild notebook v3 + clean dev-scratch files

fix(kaggle): unpin torch and loosen trl floor to prevent bnb/unsloth break

kaggle: unpin unsloth, install matched zoo pair

kaggle: refresh cell-8 promotion timing for per-phase early-stop

train: per-phase reward thresholds for early-stop

kaggle: lower convergence bar to +1.2 reward (3.1x baseline P3)

kaggle: route Patient + Nurse to 8B-instant pool

kaggle: 8B Doctor + train-until-optimal early-stop