fix: constrain NVIDIA device plugin to GPU nodes

[atharvaHJoshi]

Fixes #8858

Related to #8859.

What this PR does

The NVIDIA device-plugin DaemonSet had no nodeSelector/node affinity, so a DaemonSet
replica was scheduled on every node, including CPU-only nodes where it has no GPU
to manage — crash-looping and potentially taking down unrelated nodes via DiskPressure.

How this differs from #8859

#8859 adds a nodeSelector on nvidia.com/gpu.present: true. That label is not set
by anything in the standard eksctl flow (eksctl installs the device plugin directly; it
does not use the GPU Operator / node-feature-discovery), so under that PR the DaemonSet
would never schedule at all and GPUs would not be exposed as resources.

This implementation instead:

• derives NVIDIA-capable instance types from the eksctl cluster configuration
  (unmanaged InstanceType + InstancesDistribution.InstanceTypes, managed
  InstanceType + InstanceTypes),
• applies a required node affinity using the standard
  node.kubernetes.io/instance-type label (always populated by kubelet), and
• also excludes Fargate/hybrid/auto compute nodes via
  eks.amazonaws.com/compute-type NotIn.

This works reliably, including on mixed instance-type nodegroups where only the GPU
variants run the plugin.

Changes

• Add SetNodeAffinity to the DevicePlugin interface; no-op for Neuron/EFA.
• Implement NVIDIA-specific node affinity in NvidiaDevicePlugin.SetNodeAffinity.
• Add regression tests (unmanaged, managed, and mixed instance distributions;
  no-op for CPU-only clusters).

Validation

• go build ./...
• golangci-lint run --timeout=30m ./pkg/...
• go test ./pkg/addons/...

Pre-existing, unrelated failures in pkg/info and pkg/iam/oidc are environmental
(kubectl/cfssl missing locally) and remain unchanged.

[github-actions]

Hello atharvaHJoshi 👋 Thank you for opening a Pull Request in eksctl project. The team will review the Pull Request and aim to respond within 1-10 business days. Meanwhile, please read about the Contribution and Code of Conduct guidelines here. You can find out more information about eksctl on our website