[Feat] Support command and args on servingEngineSpec.modelSpec

[dundysm]

The main vLLM container currently hardcodes vllm serve. This adds optional command and args on each servingEngineSpec.modelSpec[] entry so startup can be wrapped the same way init containers already can (the Prometheus/LMCache mkdir case from the issue).

When command is unset, the default vllm serve invocation is unchanged. vllmConfig.extraArgs still applies only to that default path.

Fixes #682

• Make sure the code changes pass the pre-commit checks. Helm-only change; ran helm lint and helm template.
• Sign-off your commit by using -s when doing git commit
• Try to classify PRs ([Feat])

Testing

• helm lint helm/ — pass (chart dependency warning only)
• helm template with chart defaults — main container command stays vllm serve
• helm template with modelSpec.command / modelSpec.args — custom sh -c wrapper, default serve is not in command
• Added a helm unittest case in helm/tests/deployment-vllm-multi_test.yaml

[gemini-code-assist]

Code Review

This pull request introduces support for configuring custom commands and arguments for the main vLLM container in the multi-deployment Helm chart. It updates the deployment template to conditionally render command and args if specified in the model configuration, adds corresponding schema definitions in values.schema.json, provides commented-out examples in values.yaml, and includes a unit test to verify the new behavior. There are no review comments, and I have no feedback to provide.