Manage metadata refresh logic in ClusterMetadata (via attached manager)

[dpkp]

• Use long running async loop for metadata refresh
• cluster runs loop
• Move refresh/loop -> cluster (wip)
• bootstrap_async
• Fixup wakeup/notify; catch exceptions
• manager fixture
• Fixup cluster request_update calls
• use manager.send not conn.send_request in refresh_metadata
• Move TestKafkaConnectionManagerMetadataRefresh -> test_cluster.py
• use cluster fixture
• Fix test_bootstrap_retries_empty_brokers
• modify bootstrap logging
• ClusterMetadata.attach(manager)
• Unify metadata_refresh_in_progress handling
• No return from refresh_metadata
• longer timeout for request_update send test
• weakref attach; use self._manager
• Use refresh_future to track metadata in progress; protect concurrent access to refresh_metadata