fix: vertex stuck not reaching Running due to lastScaledAt null validation error after SSA migration

[Koalk]

Bug description

After PR #2570 migrated vertex status updates to Server-Side Apply (SSA), vertices can get permanently stuck in a reconcile error loop and never reach Running phase, producing no pods.

Error

The controller logs the following error on every reconcile loop:

Vertex.numaflow.numaproj.io "<vertex-name>" is invalid: status.lastScaledAt: Invalid value: "null": lastScaledAt in body must be of type string: "null"

Root cause

With SSA, the entire VertexStatus struct is serialized and sent as a JSON patch. A zero-value metav1.Time (i.e. LastScaledAt that has never been set) serializes to JSON null. The full CRD schema defines lastScaledAt as type: string, format: date-time without nullable: true, so the Kubernetes API server rejects the patch.

This creates a deadlock:

1. Vertex created → LastScaledAt is zero → SSA writes null → API rejects
2. Vertex never reaches Running
3. Autoscaler skips it: "Vertex not in Running phase, skip scaling"
4. LastScaledAt never gets set → back to step 1

The trigger condition is any vertex where currentReplicas == desiredReplicas on first reconcile (e.g. a source vertex with min: 1, max: 1, or any vertex during pipeline pause with min: 0, max: 0), because the scale branch that sets LastScaledAt is never taken.

Affected versions

All versions since #2570 was merged (v1.7.x+). The minimal CRD install is not affected since status uses x-kubernetes-preserve-unknown-fields: true.

Fix

Add // +nullable to LastScaledAt in VertexStatus and MonoVertexStatus so the generated CRD schema includes nullable: true, allowing the API server to accept null for this field.

[vigith]

would you like to create a PR for this?

[Koalk]

On it #3358