Clear logcontext before starting fed txn queue runner

These processes take a long time compared to the request, so there is lots of
"Entering|Restoring dead context" in the logs. Let's try to shut it up a bit.

synapse/federation/transaction_queue.py

@@ -20,7 +20,7 @@ from .persistence import TransactionActions
 from .units import Transaction, Edu
 
 from synapse.api.errors import HttpResponseException
-from synapse.util import logcontext
+from synapse.util import logcontext, PreserveLoggingContext
 from synapse.util.async import run_on_reactor
 from synapse.util.retryutils import NotRetryingDestination, get_retry_limiter
 from synapse.util.metrics import measure_func
@@ -146,7 +146,6 @@ class TransactionQueue(object):
         else:
             return not destination.startswith("localhost")
 
-    @defer.inlineCallbacks
     def notify_new_events(self, current_id):
         """This gets called when we have some new events we might want to
         send out to other servers.
@@ -156,6 +155,13 @@ class TransactionQueue(object):
         if self._is_processing:
             return
 
+        # fire off a processing loop in the background. It's likely it will
+        # outlast the current request, so run it in the sentinel logcontext.
+        with PreserveLoggingContext():
+            self._process_event_queue_loop()
+
+    @defer.inlineCallbacks
+    def _process_event_queue_loop(self):
         try:
             self._is_processing = True
             while True: