v2.2.0 retry still failed on T3 (rv5 flake escalation): 2-core CI runner
with parallel vitest workers couldn't surface the developer dispatch line
within 10s — the trailing human-taskbook text in the timeout dump was
normal output, not a gate rejection. Local runs (CI=true) pass, confirming
a slow-machine timing flake, not a regression. All waitText sites across
rv5/mx1 widened uniformly.