linstor: NFS server init container runs the upstream piraeus-csi image instead of the platform build #5307

The NFS server DaemonSet in the linstor package runs one init container on the upstream CSI image instead of the image this repository builds. The container is linstor-wait-node-online, and it runs quay.io/piraeusdatastore/piraeus-csi, the operator default. Every other container that uses the CSI image is pinned to linstorCSI.image in packages/system/linstor/templates/cluster.yaml. That file has no nfsServer.podTemplate, so this one is left to the operator.

The image follows the operator version, not ours. In piraeus-operator the DaemonSet declares this init container with the linstor-csi image placeholder (pkg/resources/cluster/nfs-server/nfs-server-daemon-set.yaml). The operator resolves the placeholder through its image config, which in the vendored chart points at quay.io/piraeusdatastore/piraeus-csi with the operator's own tag (packages/system/piraeus-operator/charts/piraeus/templates/config.yaml). So the tag changes with every operator update and ignores linstorCSI.image. A cluster that pulls only from a mirror of the platform images still needs quay.io for it.

The impact is low. This container only waits for the LINSTOR node to come online and does not touch the LINSTOR database. Still, it is the same kind of gap that #5166 closes for the LINSTOR server init containers: an image the package leaves unset silently follows the operator default. The same PR shows how that breaks. The operator switched the CSI client certificate to LS_USER_CERTIFICATE_FILE and LS_USER_KEY_FILE, which linstor-wait-until v1.0.1 reads and v0.3.1 does not.

I'd pin it the same way the CSI node pods are pinned. Add an nfsServer.podTemplate to the LinstorCluster in cluster.yaml that sets the linstor-wait-node-online image to {{ .Values.linstorCSI.image.repository }}:{{ .Values.linstorCSI.image.tag }}, and a helm-unittest assertion next to the existing image pins in packages/system/linstor/tests/.