Initial release: Go bindings for MLX GPU framework

Provides Go bindings for GPU-accelerated array operations:
- Metal backend for Apple Silicon (M1/M2/M3/M4)
- CUDA backend for NVIDIA GPUs
- CPU fallback with SIMD optimizations
- ONNX Runtime fallback for Windows

C++ backend: github.com/luxcpp/gpu (Apple MLX fork)

Fix package structure for CGO and non-CGO builds

- Extract shared types to types.go (no build tags)
- Add package-level functions to mlx_fallback.go
- Remove nocgo.go panic (use proper fallback instead)
- Add build tags to metal tests (darwin && cgo)

The package now:
- Builds with CGO (requires MLX library for full tests)
- Builds and tests without CGO (fallback mode with stubs)

Initial release: Go bindings for MLX GPU framework

Provides Go bindings for GPU-accelerated array operations:
- Metal backend for Apple Silicon (M1/M2/M3/M4)
- CUDA backend for NVIDIA GPUs
- CPU fallback with SIMD optimizations
- ONNX Runtime fallback for Windows

C++ backend: github.com/luxcpp/gpu (Apple MLX fork)