[Feat] Add per-model custom labels and podLabels to modelSpec

[NasAmin]

Add support for custom labels and podLabels on a per-model basis via modelSpec, similar to the existing annotations and podAnnotations fields.

labels are applied to the Deployment resource metadata, and podLabels are applied to the pod template metadata. Neither is added to selector.matchLabels since selectors are immutable after creation.

Use case: Enables teams to apply custom labels per model for GPU scheduling (e.g., nodeAffinity rules driven by labels) and observability via pod metadata collection.

Changes:

• helm/values.yaml — Added labels and podLabels fields to modelSpec
• helm/values.schema.json — Added schema definitions for both fields
• helm/templates/deployment-vllm-multi.yaml — Render per-model labels on deployment and pod template metadata
• helm/tests/deployment-vllm-multi_test.yaml — Added test verifying labels, podLabels, and selector isolation

FIX #934

[gemini-code-assist]

Code Review

This pull request introduces the ability to define per-model custom labels and pod labels in the multi-model vLLM deployment template. The changes include updates to the Helm chart templates, values schema, and default values, along with a new test case to verify that these labels are rendered correctly and do not leak into the selector match labels. The reviewer suggests reordering the label blocks in the templates to ensure that model-specific labels take precedence over standard labels by placing them after the helper includes.

[ruizhang0101]

Could you add this to readme as well? Otherwise LGTM.

[NasAmin]

Could you add this to readme as well? Otherwise LGTM.

Thanks, readme updated

[ruizhang0101]

@NasAmin Could you fix the pre-commit when you have time?