Add external worker lifecycle supervision

[superposition]

Parent: #12

Goal

Add a first runnable supervision layer for external worker processes so Leash can own process lifecycle without folding every adapter into the core runtime.

Acceptance Criteria

• Define a small external-worker spec with command, args, env, restart policy, and health metadata.
• Add a supervisor that can start workers, report running/exited state, and stop them cleanly.
• Keep defaults simulation-safe: no physical actuation, no network requirement, and no worker process started unless explicitly configured.
• Expose worker state through an existing CLI or stack/module introspection path.
• Add focused unit/smoke coverage that does not require hardware.

Notes

This should build on the existing module graph and stream-hub work rather than becoming a full distributed runtime.

[superposition]

Merged to main via PR #99: #99

Merge commit: 472095947511df6267be8be60625d50c3509769f

Local proof:

• cargo test worker::
• bash -n scripts/smoke-worker.sh scripts/smoke-all.sh
• bash scripts/smoke-worker.sh
• cargo fmt --check
• cargo run --features mcp --bin leash-schema -- --check
• git diff --check
• cargo check --all-features
• cargo test --all-targets --all-features
• cargo clippy --all-targets --all-features -- -D warnings
• cargo test --no-default-features --features sim,mcp
• scripts/smoke-all.sh
• cargo run --quiet -- worker run --name smoke-worker --hold-ms 50 -- node -e 'setInterval(() => {}, 1000)'\n- cargo package --locked --allow-dirty\n- cargo package --locked\n\nPR CI proof: all gh pr checks 99 jobs passed, including both duplicate rust jobs and every feature-matrix job.