Add Intel NPU utilization

[m-falkowski]

Hello,

This commit adds proof of concept support for the Intel NPU inference accelerator ¹²³
that is present in Meteor and Lunar Lake CPUs.

I emphasize that this is a PoC implementation to show that NPU device implementation may reuse GPU codebase.
These two are asynchronous coprocessors that expose similar statistics (utilization, VRAM/RAM memory, clocks, power stats).

For now, the NPU device is added into vector and based on is_npu_device info to generate distinct strings for GPU/NPU at the drawing stage.

To make a complete implementation there are a couple of ways I see and I am not sure which is appropriate:

1. Continue the current way but filter NPU devices from gpu-average etc.
2. Duplicate or maybe alias gpu_info into npu_info, create separate container for vector<npu_info> and reuse drawing functions,
   keeping is_npu_device in supported_functions to get the device string in a drawing.
3. Duplicate fully gpu_info and drawing utilities.

I think that the second option seems to be the most optimal and I would ask for some guidance.

Footnotes

1. NPU explained: https://intel.github.io/intel-npu-acceleration-library/npu.html ↩

2. Linux driver sources path: root/drivers/accel/ivpu/ ↩

3. NPU UMD sources: https://github.com/intel/linux-npu-driver ↩