refactor: switch InformerManager to context and cache sync lookups

[Yetkin Timocin (ytimocin)]

Description of your changes

Three small changes in pkg/utils/informer/informermanager.go:

1. Channel → context. Replace parentCh <-chan struct{} on NewInformerManager with a plain context.Context. The manager derives a child context via context.WithCancel, so Stop() still works while parent-context cancellation also drives shutdown. The ContextForChannel helper (which spawned a goroutine to bridge channel-close into context-cancel) had no remaining callers and is removed.

2. Lazy sync cache. Add a sync.Map (syncedInformers) that records GVRs whose informers have already reported HasSynced() == true. HasSynced is monotonic for the life of an informer instance, so a presence-only set with no invalidation is sufficient. The cache lets hot reconcile-path callers — pkg/utils/controller/resource_selector_resolver.go (called per CRP reconcile) and pkg/controllers/resourcechange/resourcechange_controller.go (called per resource-change event) — skip the factory mutex on subsequent lookups. The readiness probe (pkg/utils/informer/readiness) also benefits.

3. Phantom-informer guard. Add a registeredInformers sync.Map populated by AddStaticResource, CreateInformerForResource, and AddEventHandlerToInformer. IsInformerSynced consults it before calling informerFactory.ForResource(), which otherwise silently registers an unstarted, handler-less informer for any unknown GVR. Observable behaviour is unchanged (callers already treat false as "not ready, requeue"); the factory just no longer accumulates phantom entries.

Five call sites updated: cmd/hubagent/workload/setup.go, pkg/controllers/{rollout,updaterun,placement}/suite_test.go, plus the package's own test file (which now adopts t.Context() since the module is on Go 1.25).

Boy-Scout: type-name-prefix the Manager interface doc, fix include → including, fix Stop doc grammar, drop a redundant outer RLock in TestAddEventHandlerToInformer, and modernize one for i := 0; i < n; i++ loop.

Fixes #649

I have:

• Associated this change with a known KubeFleet Issue (Bug, Feature, etc).
• Run make reviewable to ensure this PR is ready for review.

How has this code been tested

• make reviewable clean (fmt, vet, golangci-lint, staticcheck, go mod tidy).
• go test -race ./pkg/utils/informer/... — all tests pass under the new signature.
• New tests:
  • TestIsInformerSynced_CachesSyncedResult — first call populates the cache after WaitForCacheSync; subsequent calls hit the cache and behaviour is unchanged.
  • TestIsInformerSynced_DoesNotCacheUnsyncedResult — a registered-but-unstarted informer reports false and is not cached, so a later sync still flips the result.
  • TestIsInformerSynced_UnregisteredGVR_ReturnsFalseWithoutCreatingInformer — unknown GVR short-circuits before factory.ForResource() and never enters either map.
• Integration test suites that consume NewInformerManager (rollout, updaterun, placement) compile and run.

Special notes for your reviewer

• Stop() semantics are preserved: it still cancels the manager's context exactly once, and cancel() of an already-cancelled context is a documented no-op.
• Both caches are per-informerManagerImpl (not global), matching the per-factory scope of the underlying informer instances.
• LoadOrStore was deliberately not used for syncedInformers — it would force-store even on HasSynced() == false, poisoning the cache. The Load then conditional Store pattern is the correct write-once-on-true idiom.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!