to be clear, i'm saying the "fork_lock" that i introduce there (used only by the SchedulerService and the DaemonPantsRunner) should be a different lock from the service lifecycle lock