fix(rbgsa-controller): guard empty selector in LWP follower roles

[sebest]

Summary

• Guard against empty status.labelSelector when building the RBGSA selector for RoleInstanceSet + LeaderWorkerPattern roles
• Without this fix, an empty base selector produces ",rbg.x-k8s.io/component-index=0" (leading comma), which breaks our materializer's /scale subresource parsing and blocks all scaling
• Add errSelectorNotReady sentinel error so the caller can distinguish the transient "selector not yet populated" case (requeue after 2s) from hard errors (propagate normally)
• Add Warning event (SelectorNotReady) for observability of the retry loop
• Clean up extractLabelSelectorDefault: propagate ctx, remove dead variable, use %w wrapping, fix typo

Test plan

• go test ./internal/controller/workloads/ -run TestExtractLabelSelectorDefault -v — 7 table-driven cases covering populated/empty/missing selector, standalone vs LWP, LWS, and workload-not-found
• go test ./internal/controller/workloads/ — all existing tests pass (no regressions)
• go vet ./internal/controller/workloads/ — clean

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[Syspretor]

@sebest LGTM, please resolve the conflicts and please also resolve the conflicts of #270 as well.

[sebest]

@sebest LGTM, please resolve the conflicts and please also resolve the conflicts of #270 as well.

conflicts are fixed on both PR.

[sebest]

@Syspretor I'm not sure the remaining failing test is caused by my PR and I cannot retry it.
let me know, thanks

[cheyang]

Review Summary

The fix is straightforward and correct: guard the empty status.labelSelector for RIS + LWP roles so we never produce a malformed leading-comma selector. The sentinel error + requeue pattern is idiomatic for transient controller conditions, and the test coverage is solid (7 table-driven cases).

Blockers before merge:

1. Merge conflict — this branch conflicts with main. Please rebase.
2. E2E failure — [v1alpha1] deploy with restartPolicy timed out (150s). The failure (workload generation not equal to observed generation) looks unrelated to your change, but CI must be green.

Code observations (non-blocking):

• The pkgerrors alias + stdlib errors split is clean.
• Dead gvk variable removal, context.TODO() → ctx, %v → %w, and the typo fix are good housekeeping.
• 2s requeue is reasonable for a condition gated on another controller's reconcile.

Once the conflict is resolved and CI passes, this looks good to go.

[cheyang]

Thanks for the fix -- the empty-selector guard and sentinel error pattern look correct in isolation, and the test coverage is solid (7 table-driven cases covering both transient and hard error paths). However the PR currently has merge conflicts with main (specifically #281 which removed the Workload field in favor of GetWorkloadSpec()), and the e2e-test is failing. Could you rebase onto the latest main, adopt role.GetWorkloadSpec(), and verify the e2e suite passes? Happy to re-review once that is done.