Apple's MLX adding CUDA support

A CUDA backend is being developed for MLX, Apple's machine learning framework. While still in early stages, the tutorial example is already functional. This addition aims to leverage CUDA's unified memory support and enable seamless development on Macs with deployment to NVIDIA-powered supercomputers. The project, sponsored by Apple, is currently tested on Ubuntu 22.04 with CUDA 11.6, though other environments may work. This development could significantly enhance MLX's versatility and appeal to researchers and developers in high-performance computing environments.