Optimize costs on cluster hibernation

[unmarshall]

How to categorize this issue?

/area control-plane
/area cost
/kind enhancement

What would you like to be added:
When a k8s cluster is hibernated then the statefulset is scaled down to 0. What remains are the expensive volumes that were mounted to each of the etcd member pods. In gardener use case this is achieved via hibernation of the shoot resource.
As discussed previously we should do the following upon identifying that an etcd-cluster needs to be hibernated:

1. Take a full snapshot and upload the same to the backing-store if the backup has been enabled.
2. Remove the PVCs. This can be done in 2 ways:
   i) Update STS and set PersistentVolumeClaimRetentionPolicy (whenScaled: Delete)and this is then followed by scaling down the STS.
   ii) Scale down STS and then explicitly delete the PVCs

Why is this needed:
Incurring cost for all expensive etcd member volumes for a cluster which has been hibernated is unnecessary. We can achieve cost savings by removing the PVCs while ensuring that these are only removed once a successful full snapshot is taken so that recovering from the last state becomes possible thus eliminating data loss.