experiment: adjusts to migrate from application to app.kubernetes.io/managed-by label

[tcondeixa]

This is an experiment to allow to migrate from the application to app.kubernetes.io/managed-by label.

Phase 1: cluster_labels: application:spilo,app.kubernetes.io/managed-by:spilo

• needs pod relabeling before replace ( current PR)
• if relabeling partially fails → we abort before deleting the old StatefulSet → pods still match the old application:spilo selector → still fully managed → no split-brain risk
• worst case: nothing changes, operator logs a warning, you retry

Phase 2: cluster_labels: app.kubernetes.io/managed-by:spilo

• Existing selector: {application:spilo, app.kubernetes.io/managed-by:spilo, cluster-name:x}
• Desired selector: {app.kubernetes.io/managed-by:spilo, cluster-name:x}
• MapContains(existing, desired) → true → already works with existing code, no fix needed
• Pods already carry owner:spilo from Phase 1 → new StatefulSet adopts them immediately
• No relabeling step, no risk window

So two-phase concentrates the risk only in Phase 1, and Phase 1's risk is bounded because the old selector still exists on pods if anything goes wrong. Phase 2 is completely safe with zero code changes.

The code change needed is only in Phase 1: relabel pods with the new labels before doing the cascade=orphan replace, and abort if any pod fails to relabel.