Feature/v0.3.0/hardware gpu support (CANCELED)

[mavdol]

hardware GPU support

This PR cancels the planned GPU/VRAM support via WASI-NN and updates the roadmap accordingly.

• Feature canceled: GPU support via WASI-NN After investigation, WASI-NN integration introduces significant constraints:

  • Large bundle size : ONNX Runtime adds ~150MB, PyTorch adds 2-3GB
  • Backend dependency: End users would need PyTorch, OpenVINO, or ONNX Runtime installed
  • Model format constraints: Users would need to convert models to ONNX format

• Roadmap updated v0.3.0 is now focused on Data Access (filesystem mounting) rather than GPU support. GPU/ML workloads can still be handled via external services (Ollama, vLLM, etc.)