chore: Add a test to validate that Terminated message handling in Cluster Sharding

[JustinPihony]

A race condition was tested with a local reproducer re #32766 , but this should be part of the regular testing cadence.

The first Terminated signal is handled specially in #32756, which actually ends being put behind other stashed messages, delaying its processing.

To reproduce required at least 9 nodes, with 3 being terminated. But the main part to reproduce was probably having ongoing shard allocation when the nodes were being terminated, so that the terminated signals were received during state update.

[JustinPihony]

https://github.com/akka/akka/blob/main/akka-cluster-sharding/src/multi-jvm/scala/akka/cluster/sharding/ClusterShardCoordinatorDowning2Spec.scala was suggested as a starting point. It doesn't test continuous activity on the shard coordinator, but waits for the cluster changes first - so we would need that as part of the test.