We should use `cycle_time` to decide when to set the deadline, instead of `max_run_time` This lets us rerun coverage tasks when they fail without retriggering the whole pool