feat(k8s): nvidia device-plugin manifest SoR — GPU verified E2E [#763][#769]

k8s now in the O&M lane (Charles 09-04). Verification: capacity/
allocatable nvidia.com/gpu=1 on wnode-tsys3; CUDA pod scheduled and
nvidia-smi clean in-container. Q2 sweep: zero node_exporters on all
PVE hosts (Beszel owns host metrics).

https://projects.knownelement.com/issues/763

k8s/code/nvidia-device-plugin.yaml

@@ -0,0 +1,34 @@
+# nvidia GPU device-plugin DaemonSet — pfv-k8s-wnode-tsys3 (Quadro M2000M 4G)
+# GPU verified end-to-end 2026-09-04 (#763/#769): pod with nvidia.com/gpu:1
+# schedules + nvidia-smi works in-container (driver 550.163.01, CUDA 12.4).
+# Prereq on the node: nvidia driver + nvidia-container-runtime (k3s).
+apiVersion: apps/v1
+kind: DaemonSet
+metadata:
+  name: nvidia-device-plugin-daemonset
+  namespace: kube-system
+spec:
+  selector:
+    matchLabels:
+      name: nvidia-device-plugin-ds
+  template:
+    metadata:
+      labels:
+        name: nvidia-device-plugin-ds
+    spec:
+      nodeSelector:
+        kubernetes.io/hostname: pfv-k8s-wnode-tsys3
+      tolerations:
+        - key: CriticalAddonsOnly
+          operator: Exists
+      containers:
+        - name: nvidia-device-plugin-ctr
+          image: nvcr.io/nvidia/k8s-device-plugin:v0.17.0
+          args: ["--fail-on-init-error=false"]
+          volumeMounts:
+            - name: device-plugin
+              mountPath: /var/lib/kubelet/device-plugins
+      volumes:
+        - name: device-plugin
+          hostPath:
+            path: /var/lib/kubelet/device-plugins