MongoD Storage watchdog should treat unexpected interruption as process-fatal

The storage engine watchdog treats interruption of one of its periodic threads as fatal to that thread, but not to the process. However, it does not start a replacement thread, meaning that such an interruption (which a sufficiently powerful user can cause with a `killOp` operation) would stop monitoring, leaving behind just an error-level log message. We should change that message to severe-level (F) and terminate the process when we encounter it.