feat: implement dataset loader, environment, and GRPO training pipeline for undertrial bail prediction

Fix 5 bugs: inference mode reset, step_counts in curriculum, adapter-only save (x3), DEMO001 false defence claim, episode_id in /reset

Fix all audit gaps: custody neutral, parity-first bias, skip penalty 0.40, statutory process reward, /observation endpoint, reset() timeout, drift determinism

Fix demo: remove shuffle for deterministic seed=0; update stage 2+4 memos to match real seed=0 GT outcomes

Add seed param to /reset: demo pins to seed=0 per stage for consistent known episodes

Add 4 rich demo episodes (all stages incl. bias+BNSS); demo UI cycles random stages

first commit