fix(kura): reconcile expanded storage without StatefulSet mutation

What changed

• Preserve existing StatefulSet volumeClaimTemplates during KuraInstance reconciliation so storage changes do not attempt to mutate an immutable StatefulSet field.
• Reconcile the actual data-<instance>-<ordinal> PVC requests up to the desired spec.storageSize and skip missing PVCs until the StatefulSet creates them.
• Refuse to shrink PVCs when the desired size is lower than the current request.
• Grant the Kura controller RBAC to read and patch PVCs.
• Raise the standalone Kura Helm chart persistence default to 200Gi, matching the controller default for new Kura instances.

Why

Existing StatefulSets can have older PVC templates even after their PVCs are expanded. Rewriting volumeClaimTemplates on every reconcile causes Kubernetes to reject otherwise valid updates, which blocks image rollouts and normal controller convergence.

The safer path is to treat the StatefulSet template as immutable once it exists, then reconcile live PVCs directly for supported expansion. New StatefulSets still get the desired storage size from the template.

Impact

Kura instances can converge to 200Gi storage without deleting StatefulSets or orphaning volumes. Existing pods keep their retained cache volumes, and future scale-ups are corrected by the PVC reconciliation loop if their PVC is created from an older template.

Validation

• git diff --check
• mise exec go@1.25.0 -- go test ./... from infra/kura-controller