Move DB manager exit check to happen in main thread (#3922)

Prior to this PR, self.close() was called inside the message migration
thread. But self.close() does database stuff, which should all happen in
one thread: the main thread.

The historical reason is that self.close was originally triggered by a
special stop message in what was previously many threads all running
this migrate code. That stop message was removed in PR #3807 and
replaced by a multiprocessing Event.

This addresses some exceptions at shutdown when both the main thread and
the message migration close() attempt to write to the database at the
same time.

This PR does not address the possibility of outstanding messages which
have not reached the database manager at the time of shutdown.

The message migration thread now does only message routing.

## Type of change

- Bug fix

Author: benclifford

parsl/monitoring/db_manager.py

@@ -547,6 +547,10 @@ def start(self,
                     "or some other error. monitoring data may have been lost"
                 )
                 exception_happened = True
+
+            if self.external_exit_event.is_set():
+                self.close()
+
         if exception_happened:
             raise RuntimeError("An exception happened sometime during database processing and should have been logged in database_manager.log")
 
@@ -558,9 +562,6 @@ def _migrate_logs_to_internal(self, logs_queue: mpq.Queue, kill_event: threading
             logger.debug("Checking STOP conditions: kill event: %s, queue has entries: %s",
                          kill_event.is_set(), logs_queue.qsize() != 0)
 
-            if self.external_exit_event.is_set():
-                self.close()
-
             try:
                 x = logs_queue.get(timeout=0.1)
             except queue.Empty: