src: fix platform shutdown deadlock

Each worker is signalling its own completion of tasks independently
and so should only be signalling for one corresponding drain otherwise
the count of outstanding tasks goes out of sync and the process will
never stop waiting for tasks when it should be exiting.

It just needs to be calling Signal rather than Broadcast.

Author: Qard

src/node_platform.cc

@@ -629,7 +629,7 @@ template <class T>
 void TaskQueue<T>::NotifyOfCompletion() {
   Mutex::ScopedLock scoped_lock(lock_);
   if (--outstanding_tasks_ == 0) {
-    tasks_drained_.Broadcast(scoped_lock);
+    tasks_drained_.Signal(scoped_lock);
   }
 }