Quickfix for #384: Don't defrag etcd on leader.

Created by: garloff

This is the quick-fix: We just don't do defragmentation on the etcd leader. This avoids etcd service interruption on single-node etcd clusters and spurious leader changes on multi-node ones.

Note that this is the intermediate step until we have a more complete solution as depicted in https://github.com/SovereignCloudStack/k8s-cluster-api-provider/issues/384#issuecomment-1468290695