In certain cases, a race condition seems to come up where the PID
retrieval thread tries to set the PIDs while the cgroup object is being
deleted. So lock the mutex in the destructor, so that if that should
happen, destruction will block while the PIDs are being updated.