test: cover LWS multi-vendor GPU resources

[xuhui-lu]

Summary

• add LWS GPU counting coverage for supported resource names: NVIDIA, AMD, Intel Gaudi, Intel i915, and Intel Xe
• add a mixed-vendor LWS case across leader and worker containers
• keep this as a unit-test-only follow-up for the multi-vendor GPU work

Part of #1106.

Testing

• go test ./internal/utils/scaletarget
• go test ./internal/utils/...

Also tried go test ./...; it is not clean locally for unrelated reasons:

• internal/engines/analyzers/throughput: FitITLModel flat-line test currently fails
• test/chart: local helm binary is not installed
• test/e2e: entered cluster preflight / auth flow and was interrupted

[Copilot]

Pull request overview

This PR extends unit-test coverage for LWSAccessor.GetTotalGPUsPerReplica() to ensure GPU counting works across the full set of supported vendor resource names (NVIDIA, AMD, Intel Gaudi, Intel i915, Intel Xe), including a mixed-vendor leader/worker scenario. This supports the broader multi-vendor GPU support work tracked in #1106 by guarding against regressions in GPU resource parsing for LeaderWorkerSet scale targets.

Changes:

• Added a table-driven unit test covering all supported GPU resource names for LWS GPU counting.
• Added a mixed-vendor LWS test case that combines different vendor resources across leader and worker containers.
• Introduced small test helpers to reduce duplication when building LWS/container GPU requests.

[ev-shindin]

LGTM. Look to some minor (non-blocking) comments

[ev-shindin]

/ok-to-test

[github-actions]

🚀 Kind E2E (full) triggered by /ok-to-test

View the Kind E2E workflow run

[github-actions]

🚀 OpenShift E2E — approve and run (/ok-to-test)

View the OpenShift E2E workflow run

[github-actions]

GPU Pre-flight Check ✅

GPUs are available for e2e-openshift tests. Proceeding with deployment.

Resource	Total	Allocated	Available
GPUs	50	27	23

Cluster	Value
Nodes	16 (7 with GPUs)
Total CPU	993 cores
Total Memory	10383 Gi
GPUs required	4 (min) / 6 (recommended)

[ev-shindin]

Thanks @xuhui-lu ! To follow-up: every existing and new test sets a non-zero GPU request, so the if total == 0 { return 1 } default is never exercised — change it to return 0 and the whole suite still passes. Add a CPU-only (no GPU requests) LWS case asserting 1