ggml_cuda: the CUDA backend of llama.cpp b10786 with the runtime libraries it imports, and the licence

README: what a backend folder is and how it is used

initial commit