Avoid playing queue pollution when restoring Sonos snapshots (#21963)

Assume a snapshot state with three speakers in two groups, AB and C. They will
be playing the A and C queues, respectively. The B queue exists but is hidden
in this topology.

Unjoin B and form a new group BC, playing the B queue (now with the C queue
hidden).

To restore the snapshot we would join B back to A. The BC group would now only
contain the C speaker, still playing the B queue. The C queue has been lost :-(

The problem is that unjoining a coordinator will elect a new coordinator that
inherits the group queue and thus has its hidden queue overwritten.

This commit avoids the situation by having restore unjoin all slaves. Above, C
would be unjoined before joining B to A. This restores the C queue and since B
is then alone, it can be joined to A without having to transfer its playing
queue to remaining speakers.

homeassistant/components/sonos/media_player.py

@@ -1037,8 +1037,13 @@ def restore_multi(entities, with_group):
             if entity.state == STATE_PLAYING:
                 entity.media_pause()
 
-        # Bring back the original group topology
         if with_group:
+            # Unjoin slaves that are not already in their target group
+            for entity in [e for e in entities if not e.is_coordinator]:
+                if entity._snapshot_group != entity._sonos_group:
+                    entity.unjoin()
+
+            # Bring back the original group topology
             for entity in (e for e in entities if e._snapshot_group):
                 if entity._snapshot_group[0] == entity:
                     entity.join(entity._snapshot_group)