docker : add git to full-cuda.Dockerfile main-cuda.Dockerfile

[polym]

Add git to full-cuda.Dockerfile main-cuda.Dockerfile which is required in scripts/build.sh.

[polym]

@canardleteer PTAL

[canardleteer]

LGTM. It appears that scripts/build-info.sh was added sometime in May, adding a dependency on git, which is why these CUDA Dockerfile's broke.

This, and similar breakage, will continue to happen until there are CUDA workers available for this projects CI validation automation, since it's not on the automated testing path.

Given that there's significant CUDA usage in llama.cpp now, it may be a worthy effort for this project to attempt to go and try to find a donation of some. I don't currently have the time to volunteer for this effort (but wish I did!).

• Managed GitHub GPU runners are on the GitHub Roadmap.
• But since that's incomplete, we'd probably need to find someone offering hosting & manage it until it becomes available through GitHub.

Thanks @polym!

[ggerganov]

I have already setup an Azure-based CI with a lot of credits available (thanks to AI Grant).
There is currently 1 GPU runner on master: ggml-org / ggml-4-x86-cuda-v100

We can add more and improve the CI with more tests

[canardleteer]

We can add more and improve the CI with more tests

Good to know @ggerganov! When I get a chance, I'll see if I can put something together that can work with this setup.