Avoid failing snapshots due to transient network failures

Currently, when the data distributor sends a snapshot request to a stateful process, it can take many seconds for the stateful process to reply to the `*SnapRequest` message (depending on the underlying snapshot procedure for the stateful process). If the connection between the data distributor and is broken during this period, the entire snapshot will fail. Ideally, we should instead attempt to reestablish a connection.