More resilient Shrink Action in ILM to node dropping from cluster.

[talevy]

The ILM Shrink Action sets the allocation settings for the index to a specific node by _id. If the node with this allocation disconnects and never returns, then the index's allocation setting index.routing.allocation.require._id will never be valid and the shard will never be allocated correctly. There needs to be a more resilient step after the SetSingleNodeAllocateStep that verifies the node still exists and retries if the node is no longer part of the cluster.

[elasticmachine]

Pinging @elastic/es-core-features (:Core/Features/ILM+SLM)

[sophiaxu8]

Is this #41170 fix this issue?

[dakrone]

@sophiaxu8 no, that change was related to internal tests. We're still working on the resiliency aspect for interrupted shrink ILM operations

[dakrone]

Closing this in favor of implementing #63519