Status: IN
`Coordinator.recover()` scans the log once and re-sends decisions to currently-available participants; it does not retry or schedule follow-ups, so a participant still down at recovery time stays locked until `recover()` is manually called again.