jgrusewski
6e0c9abff2
feat(rl): regime_observer trainer wiring — loaders + struct + launchers (F1.4)
Adds:
- include_bytes! for rl_regime_flat_count + rl_regime_observer cubins
- IntegratedTrainer struct fields: 2 modules + 2 functions + flat_count_d helper buffer
- new() cubin loading + struct field assignment
- launch_rl_regime_flat_count(): Grid=(1), Block=(256), smem=256*sizeof(int)
- launch_rl_regime_observer(): Grid=(1), Block=(1), smem=0
Does NOT wire launchers into step_with_lobsim (that's F1.5).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>