feat: Intel Hardware Support

[tikikun]

Currently nitro can be compiled with intel MKL, however there are ongoing efforts from intel for LLMs inferencing that offers better performance.

• Hugginface Optimum with OpenVINO https://huggingface.co/docs/optimum/intel/inference
• BigDL https://github.com/intel-analytics/BigDL
• Intel extensions for transformer https://github.com/intel/intel-extension-for-transformers

Criteria that we might want:

• Ability to be statically compiled
• As few dependencies as possible (if none will be best like llama cpp)

That said with above options and multiple documents regarding on how to do things on intel hardware i think it will be awesome if we can have discussion and more insights on this matter.

[tikikun]

regarding issue that leads to this discussion:

• bug: 1 token after 5 minutes for Jan on Windows without GPU jan#544