Today submitting an unbatched ClusterStateUpdateTask to the MasterService is deprecated, and devs are encouraged to write their own batching executor instead. In principle this makes it possible to combine updates without incurring the costs of building a whole new ClusterState on each iteration, but in practice most implementations don't work at a lower level and just iteratively update a ClusterState. Moreover creating new ClusterState instances is nothing like as expensive as once it was.

We should accept this and provide some friendlier utilities for writing batched cluster state updates, ideally leaving the developer to express the desired per-task update as a ClusterState -> ClusterState operator.