Use client-go / apimachinery label selector utilities instead of custom label matching helpers in the PDB controller.

[ljluestc]

Summary

Use client-go / apimachinery label selector utilities instead of custom label matching helpers in the PDB controller.

Fixes #8.

---

Background / Motivation

The controller previously implemented its own label matching helpers (containLabels, labelsIntersect) to:

• Decide which PDBs are managed vs unmanaged (based on heritage: pdb-controller).
• Determine whether existing unmanaged PDBs already match a workload’s pod template labels (getMatchedPDBs).

Kubernetes already provides robust label selector types and matching logic via client-go / k8s.io/apimachinery, so this PR replaces our ad‑hoc implementations with those built‑in utilities. This reduces duplicated logic and keeps behavior consistent with other Kubernetes components.

---

What this PR changes

• Replace custom intersection logic in getMatchedPDBs

  • Before: getMatchedPDBs relied on a custom labelsIntersect function comparing two map[string]string.
  • After: it now uses Kubernetes’ label selector APIs:
    • Convert each PDB’s Spec.Selector (*metav1.LabelSelector) using metav1.LabelSelectorAsSelector.
    • Convert the resource’s pod template labels to a labels.Set.
    • Use selector.Matches(labels.Set(resourceLabels)) to decide whether the PDB matches the workload.

• Use label selectors to find managed PDBs

  • Before: filterPDBs used containLabels(pdb.Labels, ownerLabels) to decide whether a PDB is managed.
  • After: filterPDBs uses labels.SelectorFromSet(ownerLabels).Matches(labels.Set(pdb.Labels)), delegating the match semantics to apimachinery.

• Remove custom helpers that duplicate selector behavior

  • labelsIntersect and the custom subset logic in containLabels are no longer used for controller behavior.
  • Managed/unmanaged classification and matching now go through the selector utilities, which are well‑tested and idiomatic in Kubernetes code.

• Tests

  • Existing behavior tests in controller_test.go continue to exercise:
    • Classification of managed vs unmanaged PDBs.
    • Matching between workloads and PDBs (including the parent‑resource‑hash paths).
  • No changes to the public API or configuration flags.

---

Behavior notes

• Matching semantics now exactly follow Kubernetes label selector behavior:
  • PDB selectors can use the standard LabelSelector form (we already use MatchLabels, but this prepares us for more complex selectors).
  • Any invalid LabelSelector will be safely skipped (logged and ignored) rather than leading to undefined behavior.
• Managed PDB detection still uses the same ownerLabels (heritage: pdb-controller); only the implementation is switched to selectors.