[P1][Post-0.5][Qualification] Automate colocated and remote lifecycle conformance with failure injection

[mightnent]

Parent qualification epic: #79

Outcome

Run one topology-neutral workspace lifecycle conformance suite across colocated and remote-node paths, with deterministic failure injection and cleanup.

Execution

• Priority: P1 post-0.5 qualification.
• Deployment profiles: Worktree colocated and remote required; customer-managed/hosted evidence where applicable.
• Worktree boundary: Lifecycle fixtures, adapter/controller/Control orchestration, fault injection, reconciliation assertions, and evidence output. Avoid implementing Full C placement or recovery policy.

Scope

• Exercise create, readiness/status, open, reconnect, live policy/egress update, stop, restart, persistence reuse, destroy, and generation-fenced purge through the same contract in both topologies.
• Verify both paths use the same immutable workspace-image digest.
• Inject delayed and malformed readiness, terminal failure, timeout, partial create, stop failure, controller restart, provider drift, missing runtime, orphan runtime/relay, and removal-in-progress races.
• Prove deterministic retry, reconciliation, non-overlapping lifecycle mutations, and qualification-owned cleanup.
• Record phase timings, typed failure codes, topology, allocation, home size/inodes, SHA, and image digest without secrets.
• Allow Full C implementation issues under [P1][Post-0.5][Epic] Productionize the Full C multi-node workspace runtime #82 to add multi-node, stale-generation, drain, and recovery scenarios without forking the lifecycle suite.

Non-goals

• Fixing the data-bearing startup regression owned by [P0][0.5] Make data-bearing workspace startup reliable without whole-home repair #84.
• Browser security/network probes or detailed performance distributions.
• Implementing scheduling, provisioning, or migration behavior.

Definition of success

• Colocated and remote paths pass the same lifecycle contract with the same immutable image digest.
• Every named failure produces a deterministic typed outcome and cleanup/reconciliation state.
• Controller restart, repeated request, drift, orphan, and Docker removal races do not leave a stuck lifecycle record or unmanaged runtime.
• Existing valid workspace data and unrelated persistent volumes are preserved.
• Evidence is secret-free and bound to the exact candidate SHA, topology, image, allocation, and fixture.
• Focused lifecycle qualification and npm run verify:quick pass on the exact evidence SHA.