Improve node selection logic when scaling down

As per subject, improve logic when determining which nodes to remove when scaling down cluster to match target node count, in order to reduce any unnecessary work.

• Server Version: Potentially if a previous upgrade has been interrupted, there may be extra nodes in the cluster, which may already match the target (upgrade) version, so these should be prioritised for retention
• Data Size: If the overall data size for any services that need to be rebalanced can be determined, prioritise removing nodes that have the least amount of data on them

Regarding node health - there shouldn't be any unhealthy Pods in the cluster (as these should already be removed), but we are looking to make improvements to the in-product health reporting, so it's worth checking when it comes to working this improvement in case there are any new options beyond "Pod Health".