Update build flag for consistency with llama.cpp

[jahfer]

👋🏼 About a month ago, llama.cpp updated their CUDA build command to LLAMA_CUDA=1 make rather than the prior LLAMA_CUBLAS=1 make (ggml-org/llama.cpp#6299).

This PR adds --with-cuda as a build option to the extconf.rb, while maintaining the previous --with-cublas flag to prevent breaking existing users.

[yoshoku]

@jahfer
Thank you for your contribution. This project adopts the conventional commits, please amend the commit message to something like "chore: update flags to stay consistent with llama.cpp".
https://github.com/conventional-changelog/commitlint/tree/master/%40commitlint/config-conventional

[jahfer]

Oops, thanks for the details. I've pushed the amended commit.