Test and improve intel support

[eppane]

Currently TorchServe has support for Intel GPUs:

examples/intel_extension_for_pytorch

• Support is provided more as an extension rather than being inherent part of TorchServe

• Nevertheless, TorchServe with Intel® Extension for PyTorch* can deploy any model and do inference.

• To use TorchServe with Intel GPU, the machine must have the latest oneAPI Base Kit installed, activated, and ipex GPU installed.

• There is a custom intel_gpu_metric_collector.py provided that needs to be used, rather than the existing metric_collector.py in TorchServe

Action points:

• Integrate Intel support as an inherent part of TorchServe
• Integrate the provided custom metric collector script to TorchServe
  • See also print_env_info.py

[eppane]

Some notes:
torch.xpu — PyTorch 2.5 documentation
torch.xpu.is_available — PyTorch 2.5 documentation
Intel® Extension for PyTorch* — PyTorch Tutorials 2.5.0+cu124 documentation
GitHub - intel/intel-extension-for-pytorch at xpu-main
Examples — Intel&#174 Extension for PyTorch* 2.3.110+xpu documentation
pytorch/torch/xpu/memory.py at main · pytorch/pytorch
Tensor Attributes — PyTorch 2.5 documentation