Avoid race condition in event_reactor test (#572)

These tests essentially queue an event, wait for the event reactor to
process the event, and check that an appropriate function was or was not
called. The way we were previously waiting for the event reactor thread
to do the work left a window between popping the work from the queue and
actually passing the event to the observers. If the flush loop finished
early in this window, there was a chance that the test's assert would
beat the event reactor, which hadn't yet invoked the observers.

Rather than change the behavior of the flush function to also ensure
that the popped work had completed processing, we can simply use the
queue's integrated work tracking, which won't unblock until there is no
work in the queue OR processing.

test/test_event_reactor.py

@@ -66,15 +66,15 @@ def test_create_event_reactor():
             assert error.call_count == 0
 
             queue.put(('first', None))
-            event_reactor.flush()
+            queue.join()
             assert error.call_count == 1
 
             queue.put(('second', None))
-            event_reactor.flush()
+            queue.join()
             assert error.call_count == 1
 
             queue.put(('third', None))
-            event_reactor.flush()
+            queue.join()
             assert error.call_count == 2
 
         # 1 timer event, 3 mock string events