Consistency for how verify_request.deferred is called

Define that it is run with no log context, and make sure that happens.

If we aren't careful to reset the logcontext, we can't bung the deferreds into
defer.gatherResults etc. We don't actually do that directly, but we *do*
resolve other deferreds from affected callbacks (notably the server_to_deferred
map in _start_key_lookups), and those *do* get passed into
defer.gatherResults. It turns out that this way ends up being least confusing.

synapse/crypto/keyring.py

@@ -57,7 +57,8 @@ Attributes:
     json_object(dict): The JSON object to verify.
     deferred(twisted.internet.defer.Deferred):
         A deferred (server_name, key_id, verify_key) tuple that resolves when
-        a verify key has been fetched
+        a verify key has been fetched. The deferreds' callbacks are run with no
+        logcontext.
 """
 
 
@@ -284,6 +285,7 @@ class Keyring(object):
                     if not missing_keys:
                         break
 
+                with PreserveLoggingContext():
                     for verify_request in requests_missing_keys.values():
                         verify_request.deferred.errback(SynapseError(
                             401,
@@ -294,6 +296,7 @@ class Keyring(object):
                         ))
 
         def on_err(err):
+            with PreserveLoggingContext():
                 for verify_request in verify_requests:
                     if not verify_request.deferred.called:
                         verify_request.deferred.errback(err)
@@ -714,6 +717,7 @@ class Keyring(object):
 def _handle_key_deferred(verify_request):
     server_name = verify_request.server_name
     try:
+        with PreserveLoggingContext():
             _, key_id, verify_key = yield verify_request.deferred
     except IOError as e:
         logger.warn(