The cluster command for merge chunks currently relies on possible Stale metadata to check if the user can run the command.
This is just a best-effort approach to prevent the operation from executing but it can be confusing for the user in case the operation does not execute due to stale informations

Those checks should be moved on the csrs shard