Add vLLM OCI artifact format support with media type application/vnd.docker.ai.vllm.v1

[Copilot]

This PR adds support for vLLM models as a new OCI artifact format alongside the existing GGUF support. The implementation introduces the media type application/vnd.docker.ai.vllm.v1 for vLLM models and enables seamless packaging and distribution of both GGUF and vLLM formats.

Changes Made

New OCI Media Type

• Added MediaTypeVLLM constant with value application/vnd.docker.ai.vllm.v1
• Added FormatVLLM constant for vLLM format identification

vLLM Support Infrastructure

• Created internal/vllm package with complete model creation and handling logic
• Added FromVLLM() function to the builder package for creating vLLM model artifacts
• Created Dockerfile.vllm for containerizing vLLM models

Enhanced CLI Tool

The CLI tool now automatically detects model format based on file extensions:

• .gguf files → GGUF format (existing behavior)
• .safetensors and .bin files → vLLM format (new)
• Unknown extensions → defaults to GGUF with warning (backward compatibility)

Updated Interfaces

Extended existing interfaces to support both formats:

• Added VLLMPaths() method to Model interface
• Added VLLMPath() method to ModelBundle interface
• Updated bundle unpacking logic to handle vLLM model files

Examples

Packaging a vLLM model:

model-distribution-tool package ./model.safetensors registry.example.com/models/llama:v1.0

Packaging a GGUF model (unchanged):

model-distribution-tool package ./model.gguf registry.example.com/models/llama:v1.0

The tool will automatically detect the format and use the appropriate media type in the OCI manifest.

Testing

• Added comprehensive test coverage for vLLM functionality
• All existing tests continue to pass (no breaking changes)
• Manual testing verified both formats work correctly
• Verified OCI manifests contain correct media types

Backward Compatibility

This is a purely additive change that maintains full backward compatibility. Existing GGUF workflows continue to work exactly as before, while new vLLM workflows are now supported alongside them.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.