Resolve potential deadlock in local_worker_manager (#108)

- acquire lock -> future.set_result -> done_callback() -> ... -> _Stub. __getattr__ -> remote_call -> fetch lock -> deadlocked
- Releasing the lock earlier resolves this.

Author: Northbadge

compiler_opt/distributed/local/local_worker_manager.py

@@ -161,10 +161,10 @@ def _msg_pump(self):
         with self._lock:
           future = self._map[task_result.msgid]
           del self._map[task_result.msgid]
-          if task_result.success:
-            future.set_result(task_result.value)
-          else:
-            future.set_exception(task_result.value)
+        if task_result.success:
+          future.set_result(task_result.value)
+        else:
+          future.set_exception(task_result.value)
 
       # clear out pending futures and mark ourselves as "stopped" by null-ing
       # the map