fix(aio): pin the instance for cluster-endpoint connections in aurora_connection_tracker

[AhmadMasry]

Description

Fixes #1276.

With the async API and aurora_connection_tracker (part of the default plugin chain), every connection made through a cluster endpoint was closed by the tracker on its first statement, and the app saw FailoverSuccessError.

Cause: _pin_current_writer pins the writer from topology, then its live role probe replaces the pin with plugin_service.current_host_info when the connection is a writer. For a cluster-endpoint connection that is the cluster endpoint. On the first execute, _invalidate_writer_change compares the pin with the topology writer (an instance endpoint), sees a writer change, and invalidate_all closes every connection keyed under the cluster endpoint, including the one about to execute.

Sync doesn't hit this: AuroraConnectionTrackerPlugin always takes _current_writer from topology (Utils.get_writer(all_hosts)), so it compares instance with instance.

Changes in aio/aurora_connection_tracker.py:

• _fill_instance_alias already identifies the instance a cluster-endpoint connection landed on. The plugin now keeps it as _connected_instance.
• The live-probe pin uses that instance when the connection's host isn't an instance endpoint. If it couldn't be identified (for example a custom endpoint), the topology pin is kept. This matches sync, where the writer is always a topology instance.
• _invalidate_departed_host resolves a cluster-endpoint pre_failover_host to _connected_instance. Without this, the fix above would stop the departed-host path from firing for cluster-endpoint connections when topology lags after a failover, because the pin (an instance) would no longer match the departed host (the cluster endpoint). After a move, _connected_instance follows the connection to its new instance.

Testing:

• New unit tests in tests/unit/test_aio_aurora_connection_tracker.py:
  • test_cluster_endpoint_connection_survives_its_first_execute: the pin is the instance and the first execute doesn't close the connection. Failed before the fix.
  • test_failover_invalidates_departed_instance_of_cluster_endpoint_connection: with lagged topology, a failover from a cluster-endpoint connection still closes idle connections to the demoted writer.
  • test_reader_endpoint_failover_does_not_invalidate_the_writer: a reader-endpoint connection failing over between readers doesn't close writer connections.
• Full unit suite: 2,394 passed. flake8, isort and mypy are clean on the changed files.
• Not yet run against a live Aurora cluster.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[AhmadMasry]

Superseded by #1290, which combines this fix with the related async monitoring fixes (#1284, #1276, #1287, #1288) in one PR.