TransportRolloverAction retains as many cluster state as there are rollover tasks waiting in a batch consuming too much memory #90620
Description
The way, the CS batching in TransportRolloverAction is written is broken for large batches of tasks. Each task result will contain the cluster state after applying the task. This consumes unmanageable amounts of memory for large cluster states if a larger number of rollovers gets batched, causing master nodes to run out of heap from endless RolloverResult getting retained.
We must find a way to not retain cluster states like this in the task results.