e2e: the HelmRelease history guard does not read the KubernetesNodes pool release #4737

The node pool release kubernetes-nodes-<test>-md0 is an Application HelmRelease, so the aggregated apiserver stamps RetryOnFailure on it (pkg/registry/apps/application/rest.go) the same way it stamps the parent. It sits directly on the node-join path, through the talos-reconcile Job. The status.history guard in hack/e2e-chainsaw/_lib/run-kubernetes.sh (cozy_guard_all_helmreleases) selects addon releases by the kubernetes-<test>- prefix. That prefix does not match this name, so a failed action on the pool release is never read, on the passing path or on the failure-path report. This gap predates the guard rework, since only the parent was read before it.

The fix is probably one more cozy_guard_helmrelease tenant-test kubernetes-nodes-<test>-md0 required call in cozy_guard_all_helmreleases, plus the same release in cozy_report_helmrelease_remediation, once someone confirms the release name is stable across the suites. The related csi-proxmox strategy pin is tracked in #4731.