control_connection: read actual metadata_request_timeout from Cluster.metadata_request_timeout

PR #361 that have introduced `metadata_request_timeout` intentionally
pulled actual `metadata_request_timeout` from
`control_connection_timeout`, because both `metadata_request_timeout`
and `control_connection_timeout` has to be inline.
Since this PR brings `Cluster.metadata_request_timeout` back, we need to pull it
from `Cluster.metadata_request_timeout`

Author: dkropachev

cassandra/cluster.py

@@ -3616,7 +3616,7 @@ def _try_connect(self, host):
 
         # Cassandra does not support "USING TIMEOUT"
         self._metadata_request_timeout = None if connection.features.sharding_info is None \
-            else datetime.timedelta(seconds=self._cluster.control_connection_timeout)
+            else datetime.timedelta(seconds=self._cluster.metadata_request_timeout)
         self._tablets_routing_v1 = connection.features.tablets_routing_v1
 
         # use weak references in both directions