README figure checkpoints

Trained policy checkpoints that reproduce the README figures (circularize lineage: imitation oracle → diff-sim refinement → 92.3% champion). Not committed to git (per the repo's no-blobs rule); download here and pass to the viz scripts, e.g.:

uv run --with jax --with matplotlib --with pillow python scripts/viz_readme_extra.py
uv run --with jax --with matplotlib --with pillow python scripts/viz_readme.py models/warm_r28_ema_final.npz <logdir>

• dagger_jax.npz — imitation (DAgger) oracle, 79.9% fresh-set
• warm_r19.npz, warm_r25.npz, warm_r26.npz — diff-sim refinement stages
• warm_r28_ema_final.npz — final champion, 92.3% fresh / 1.03× fuel (verified)