NetApp: release SnapMirror relationships orphaned by a DR promote

[Carthaca]

Problem

After a replica promote (DR failover), the previously active replica is
demoted and its old outgoing SnapMirror relationships are supposed to be
released by cleanup_previous_snapmirror_relationships. That release is
best-effort and is skipped when the demoted replica's host is unreachable
at promote time — delete_snapmirror sets src_client = None and returns
without releasing (and ESOURCE_IS_DIFFERENT is likewise swallowed).

Once the demoted replica converges to in_sync as a destination of the new
source, update_replica_state no longer attempts any cleanup ("Replicas
already in-sync won't try another cleanup attempt"), so the stale
relationship lingers on the backend as a broken-off orphan indefinitely.

Observed symptom: manila reports the live replica in_sync (correct — the
forward-filtered query never matches the stale reverse orphan), while
snapmirror show on the backend lists a Broken-off, Healthy=false
relationship left over from before the promote. This is not a
mis-classification of the live replica — it is an un-reconciled orphan.

Fix

• Add DataMotionSession.has_leftover_source_snapmirrors(replica, replica_list):
  lists the source-side destinations for the replica and returns True only if
  one targets another known replica volume (known-pairings only). Returns
  False when the source is unreachable (self-healing — retried next cycle) or
  on API error.
• update_replica_state now runs the existing
  cleanup_previous_snapmirror_relationships once for an in_sync replica when
  such a leftover is detected. The reported replica_state is unchanged.

This is the relationship-cleanup sibling of 7771edc ("Delete SnapMirror
snapshots left behind after a DR replica promotion"), which addressed only the
orphaned snapshots, not the orphaned relationship.

Testing

• New unit tests:
  • test_update_replica_state_in_sync_with_leftover_snapmirror
  • test_has_leftover_source_snapmirrors_found / _no_known_destination /
    _source_unreachable / _list_error
  • Updated the two existing in_sync cases to pin the new check.
• tox -e py3 for test_lib_base and test_data_motion: green.
• flake (hacking) on the changed files: clean.

AI assistance

AI generated the has_leftover_source_snapmirrors helper, the
update_replica_state gating branch, and the unit tests; human reviewed and
verified the SnapMirror direction semantics and the known-pairings safety
constraint. See the Assisted-By trailer on the commit.