Revert install-nvkind.sh to plain `go install` once nvkind/PR#76 lands

[mchmarny]

Summary

PR #1258 unblocked GPU CI by cloning nvkind at the pinned commit, sed-patching pkg/nvkind/node.go to swap --config-source=command → --config-source=file, and building from source. This was deliberately a stop-gap.

NVIDIA/nvkind PR NVIDIA/nvkind#76 ("Update nvidia-ctk --config-source flag from 'command' to 'file'", by @cdesiniotis) makes the same change upstream. Once it merges and ships in a new nvkind commit, we can drop our local patch.

Action when nvkind PR #76 merges

1. Bump .settings.yaml.testing_tools.nvkind to a commit at or after nvkind PR feat: implement Job-based validation framework with test wrapper infrastructure #76 lands.

2. Revert .github/actions/gpu-cluster-setup/install-nvkind.sh to the pre-fix(ci): build patched nvkind with --config-source=file (#1237) #1258 form:

   go install "github.com/NVIDIA/nvkind/cmd/nvkind@${NVKIND_VERSION}"
   nvkind_bin="${GOBIN:-$(go env GOPATH)/bin}/nvkind"
   "${nvkind_bin}" --help

3. Verify the three GPU jobs (smoke / inference / training) stay green on a PR that touches install-nvkind.sh.

Why we can't just wait

The guard in install-nvkind.sh greps the upstream source for --config-source=command before patching and fails CI if the string is missing. So when nvkind PR #76 merges, our next CI run will start failing on that guard — that's the trigger to do the revert, and is intentional (we never want to silently ship an unpatched build).

Related

• Issue GPU CI workflows fail: nvidia-container-toolkit 1.19.1 incompatible with kind worker containerd #1237 — original containerd schema-mismatch root cause
• PR fix(ci): build patched nvkind with --config-source=file (#1237) #1258 — the stop-gap patch this issue tracks
• Update nvidia-ctk --config-source flag from 'command' to 'file' nvkind#76 — the upstream fix this issue depends on