[TarasLykhenko]

Description of the change:

The Helm operator now sets status.observedGeneration to the custom resource's metadata.generation on every status write, following the Kubernetes API conventions.

The generation is set right after the status is read and before any condition changes. So:

• every status write (install, upgrade, reconcile, and the ReleaseFailed and Irreconcilable failure paths) records the generation it describes;
• a spec change that needs no release upgrade still produces a status write, because the stored status now differs from the new one in observedGeneration. Before this change, reflect.DeepEqual(status, originalStatus) skipped that write.

The field is additive and omitempty.

Motivation for the change:

Today the status holds only conditions and deployedRelease, so a client can't tell whether Deployed=True describes the current spec or an earlier one. After a spec update, the previous Deployed=True stays until the operator reconciles the change. If the operator is down or stuck, it stays indefinitely. A client that waits on Deployed=True before acting, for example a controller that installs a chart through a Helm-based operator, reports ready for a spec that was never applied.

With observedGeneration, a client treats the status as current only when status.observedGeneration == metadata.generation.

The same request is open against helm-operator-plugins: operator-framework/helm-operator-plugins#577.

Testing:

TestReconcileSetsObservedGeneration in internal/helm/controller/reconcile_test.go runs Reconcile with a fake client and a fake release manager for install, upgrade, a spec change with no release change, and a Sync failure. Each case checks that the stored status has observedGeneration equal to the resource's generation. All four fail without the change.

Checklist

• Add a changelog fragment
• Add tests