[WIP] ci(gcp-nvkind): run on Kubernetes 1.37 with kind v0.33.0 and nvkind main

[harche]

What type of PR is this?

/kind cleanup

What this PR does / why we need it:

Moves the GCP nvkind e2e harness (hack/ci/gcp-nvkind) from Kubernetes 1.34 to 1.37 so it can exercise newer kubelet features, such as DRA device health status (KEP-4680, ResourceHealthStatus, beta and on by default since 1.36), without extra feature gates.

Changing only K8S_VERSION isn't enough; two tooling issues block newer node images:

• kubeadm config format. kind v0.31.0, which nvkind vendors, generates a kubeadm.k8s.io/v1beta3 config. kubeadm 1.37 rejects it ("Please use kubeadm v1.36 instead"). kind v0.32.0+ generates v1beta4 for 1.36+. This PR installs kind v0.33.0 and rebuilds nvkind against it, since nvkind still vendors v0.31.0 (Bump sigs.k8s.io/kind from 0.31.0 to 0.32.0 NVIDIA/nvkind#75 is open to bump it).
• containerd runtime config. With 1.36+ images, nvkind's nvidia-ctk runtime configure --config-source=command writes a drop-in that containerd fails to restart with. Update nvidia-ctk --config-source flag from 'command' to 'file' NVIDIA/nvkind#76 switched to --config-source=file. This PR pins nvkind to c5705049 (current main), which includes it.

It also sets COPYFILE_DISABLE=1 and excludes ._* when packing the source. Without that, macOS tar adds AppleDouble files, and on a local run from a Mac, Helm fails parsing crds/._*.yaml as CRDs. Prow runs on Linux and is unaffected.

Which issue(s) this PR is related to:

N/A. Found while testing the KEP-4680 device health reporting rework (follow-up to #1243) on a real GPU.

Special notes for your reviewer:

• Prow uses the shared copy of lib/ in kubernetes/test-infra (experiment/gcp-nvkind/lib/) and sets K8S_VERSION in the job config. For the Prow job to pick this up, setup-nvkind-node.sh and K8S_VERSION need the matching change there. I'll send that once this direction is agreed.
• Validated with a local run on a GCE T4 (us-central1-b): nvkind cluster on kindest/node:v1.37.0, GPU Operator v26.3.1, driver built from source, and the Ginkgo suite passed (8/8).
• Each problem above reproduced on its own: on 1.37 with the current harness, kubeadm init fails; on 1.36.4 with the current nvkind, containerd fails to restart.
• AI disclosure: an AI coding assistant helped investigate, reproduce the runs, and draft this description.

Does this PR introduce a user-facing change?

NONE

Additional documentation (design docs, usage docs, etc.):

[kubernetes-prow]

Skipping CI for Draft Pull Request.
If you want CI signal for your change, please convert it to an actual PR.
You can still manually trigger a test run with /test all

[netlify]

✅ Deploy Preview for dra-driver-nvidia-gpu canceled.

Name	Link
🔨 Latest commit	b99ca1b
🔍 Latest deploy log	https://app.netlify.com/projects/dra-driver-nvidia-gpu/deploys/6ac3d4c2f731d7000896b35f

[kubernetes-prow]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: harche
Once this PR has been reviewed and has the lgtm label, please assign varunrsekar for approval. For more information see the Code Review Process.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[harche]

Folded into #1525.