feat(vllm): add optional --tokenizer argument for k8s deployments

[starpit]

Add support for specifying a custom tokenizer when deploying vLLM to Kubernetes.
The tokenizer argument is optional and defaults to the model's built-in tokenizer.

This implementation is for --target=k8s only. Support for --target=gce is TODO.

Changes

• Add --tokenizer CLI argument to vllm up command
• Pass tokenizer through to k8s deployment manifest
• Update deployment.yml to conditionally add --tokenizer flag to vllm serve
• Box VllmCommands and ImageCommands to fix clippy large_enum_variant warning
• Add comprehensive test coverage for tokenizer functionality

Usage

# With custom tokenizer
spnl vllm up --target k8s my-deployment \
  --model meta-llama/Llama-2-7b-hf \
  --tokenizer custom/tokenizer \
  --hf-token <token>

# Without tokenizer (uses model's default)
spnl vllm up --target k8s my-deployment \
  --model meta-llama/Llama-2-7b-hf \
  --hf-token <token>

Testing

✅ All 15 k8s tests pass
✅ Clippy checks pass with -D warnings
✅ Rustfmt formatting is correct

Made with Bob