refactor(compute): unify gateway restart reconciliation

[drew]

Summary

Remove the Docker-only gateway shutdown/startup lifecycle hooks. Gateway shutdown now leaves durable local compute alone, while startup reconciles persisted running intent through the shared idempotent ComputeDriver::StartSandbox RPC for Docker, Podman, and VM drivers.

Related Issue

Part of #2417

Changes

• remove Docker-specific shutdown cleanup and stop_managed_containers_on_shutdown
• reconcile running-intent Docker, Podman, and VM sandboxes through the compute-driver contract at gateway startup
• leave explicitly stopped sandboxes stopped and keep Kubernetes on its cluster-owned lifecycle
• add unit coverage for driver selection, phase filtering, missing resources, and start failures
• extend Docker, Podman, and VM restart E2Es to cover running and explicitly stopped intent
• document the driver-neutral shutdown/restart contract

Testing

• mise run pre-commit passes
• Unit tests added/updated
• E2E tests added/updated (if applicable)

Commands run:

• cargo test -q -p openshell-server start_persisted_sandboxes --lib — passed
• cargo test --no-run --features e2e-docker,e2e-podman,e2e-vm in e2e/rust — passed
• OPENSHELL_E2E_DOCKER_TEST=gateway_start mise run e2e:docker — passed
• OPENSHELL_E2E_PODMAN_TEST=podman_gateway_start mise run e2e:podman — not run; host has no Podman CLI
• mise run e2e:vm — blocked twice by the existing VM smoke test before the restart test (supervisor/SSH relay readiness failure)
• mise run test — all completed suites passed except the unrelated gateway_completer_returns_empty_when_no_config, which reproduces in isolation because this host has registered gateway config

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)
• Architecture docs updated (if applicable)

[github-actions]

🌿 Preview your docs: https://nvidia-preview-pr-2743.docs.buildwithfern.com/openshell