Fix runtime class cast exception in TRA

If we fail to acquire the shard lock, need to retry and wait for the new
cluster state, we were sending the wrong kind of request for the replica
action. This commit fixes this issue.

core/src/main/java/org/elasticsearch/action/support/replication/TransportReplicationAction.java

@@ -542,7 +542,7 @@ public abstract class TransportReplicationAction<
                             new TransportChannelResponseHandler<>(logger, channel, extraMessage,
                                 () -> TransportResponse.Empty.INSTANCE);
                         transportService.sendRequest(clusterService.localNode(), transportReplicaAction,
-                            new ConcreteShardRequest<>(request, targetAllocationID),
+                            new ConcreteReplicaRequest<>(request, targetAllocationID, globalCheckpoint),
                             handler);
                     }