Add workflow for tensorcore CUDA wheels

[oobabooga]

This adds a workflow for llama-cpp-python CUDA wheels with tensor cores support, and appends it to the Build Wheels for New Release with Prioritization workflow.

• The wheels are named llama_cpp_python_cuda_tensorcores.
• They are already being used in text-generation-webui. The wheels are being generated on my fork. Here is where they are imported: https://github.com/oobabooga/text-generation-webui/blob/main/modules/llamacpp_model.py#L22

I have been having tremendous difficulty in compiling those wheels, as I get flooded with errors saying

API rate limit exceeded for installation ID

that cause part of the wheels to not be uploaded to GitHub. So I end up having to run the same workflow several times for each new release.

Including this additional workflow will increase the compilation time for your actions, so I understand if you want to not include this in the repository. In this case, I can keep maintaining those additional wheels myself.

PS: MacOS wheels have been building successfully for all versions, so they can be added back if you haven't already.

[Smartappli]

Please create a batch file, and integrate your script in build-wheels-full-release.yml. Adapt powershell scripts (ps1 files)