As long as you configure things properly, yes. By 'configure things properly', basically make sure that your timeouts for the shard regions is longer than the time it takes to shut down the entities living on the node.
Also worth noting that depending on the pattern used to send to the entity, there is a possibility of dropped messages during such a case.
So you're telling me that during a network partition, Akka will somehow know that the entity actor is alive and well in (one of) the other partition(s) and won't start a new one?