That would require listening to cluster events. With the consistent hashing, when the node topology changes (nodes are removed or added) using hashing for routing will mean that the messages may start going to another actor than it was routed to before. To know what hashes went to a specific actor, you'd likely have to report back from the actor for every message. If the state is tied to specific hashes it will be problematic though, because how can you guarantee you got to restore the right state before a message was routed to the actor.
You may want to look into Akka Cluster Sharding which does routing based on an identifier, and makes sure there is only one actor alive for a given id, rebalancing when the cluster changes etc. Very useful for persistent state bound to a specific recipient.