Fix placement for malformed GPU pod requests

[googs1025]

Summary

• Keep opted-in Pods gated when they request nvidia.com/gpu without an accelerator-type label
• Avoid silently treating malformed GPU requests as CPU-only placements
• Add a placement regression test covering the malformed request

Test Plan

• /Users/zhenyu.jiang/go/pkg/mod/golang.org/toolchain@v0.0.1-go1.26.4.darwin-arm64/bin/go test ./internal/controller ./pkg/util -count=1

[Copilot]

Pull request overview

This PR changes the Pod placement controller to treat malformed GPU requests (a nvidia.com/gpu count without an accelerator-type label) as invalid and keep the Pod scheduling-gated, instead of silently placing it as CPU-only.

Changes:

• Handle util.AcceleratorRequest(pod) errors in placement selection and leave the Pod gated on invalid accelerator requests.
• Add a regression test ensuring a malformed GPU request does not get a provider nodeSelector or NodeClaim.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
internal/controller/pod_placement_helpers.go	Stops ignoring AcceleratorRequest errors so malformed GPU requests remain gated and unplaced.
internal/controller/pod_placement_controller_test.go	Adds a controller-level regression test for GPU-count-without-accelerator-type behavior.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[googs1025]

/kind bug

[kerthcet]

/lgtm
/approve

Thanks @googs1025

[InftyAI-Agent]

Approved: PR has both lgtm and approved labels