feat: support custom runtimeClass and topology namespace

[mu8086]

• Allow configuring a custom runtimeClass.name to avoid conflict with NVIDIA's default runtime
• Topology server namespace in nvidia-smi is now configurable via the TOPOLOGY_CM_NAMESPACE environment variable instead of being hardcoded to gpu-operator

Example Helm upgrade command

helm upgrade --install fake-gpu-operator ~/git/fake-gpu-operator/deploy/fake-gpu-operator \
  --namespace runai --create-namespace \
  --set runtimeClass.name=fake-nvidia

Example: Verified Pod Spec

This Pod verifies that the custom runtimeClass and dynamic topology namespace injection works correctly.

Click to expand pod.yaml

apiVersion: v1
kind: Pod
metadata:
  name: "1"
spec:
  runtimeClassName: fake-nvidia
  containers:
  - name: ubuntu
    image: ubuntu:22.04
    command: ["/bin/bash", "-c"]
    args:
      - |
        sleep infinity;
    resources:
      limits:
        nvidia.com/gpu: 1
    env:
      - name: NODE_NAME
        valueFrom:
          fieldRef:
            fieldPath: spec.nodeName