refactor(imitation): isolate LeRobot policy runtime

[TomCC7]

This is layer 2 of the OpenYAM learning stack. It introduces the isolated LeRobot policy runtime and its typed DimOS module boundary without collection or rollout orchestration.

Stack

1. feat: add isolated Python module runtime #3478 — external Python native-module runtime
2. fix(openyam): align learning stack with canonical model #3853 — OpenYAM learning baseline
3. refactor(imitation): isolate LeRobot policy runtime #3856 — isolated LeRobot policy runtime
4. feat(imitation): add native OpenYAM collection #3854 — native OpenYAM collection
5. feat(imitation): add controlled OpenYAM policy rollout #3855 — controlled OpenYAM policy rollout

Existing review history remains on #3315 because GitHub prevents changing the base branch of a PR already contained in a larger Stack. This PR points at the same tested policy commit with the correct narrow base.

Validation: 21 host-side and 14 isolated-runtime tests passed.

[TomCC7]

Superseded by #3315 after the split baseline, policy, collection, and rollout layers were inserted into the original stack.