This MR adds a Dockerfile so that a CUDA-enabled image is built every time we push to the main branch or tag.