What This PR Does

Adds Intel GPU (XPU) support to the Unsloth studio section in the README.
Depends on #9084 which enables XPU auto-detection and llama.cpp SYCL compilation, which this PR also validates it on Intel hardware, see section below.

Validation

Tested #9084 implementation and verified its working on an Intel 4x Arc Pro B60 24GB (XPU).
I had to explicitly source oneapi/setvars.sh to enable sycl-ls and icpx on PATH and set UNSLOTH_LLAMA_FORCE_COMPILE=1 (from studio/setup.sh) to force the source build path so that llama.cpp builds on the SYCL backend rather than Vulkan backend.

Install command for Intel XPU detection:

source intel/oneapi/setvars.sh    # intel oneapi path
cd unsloth && UNSLOTH_LLAMA_FORCE_COMPILE=1 . install.sh     # with PR9084 changes

Note: Not sourcing oneapi/setvars.sh also works. When sycl-ls/xpu-smi are not on PATH, it fallbacks to checking for vendor 0x8086 in /sys/bus/pci/devices under the _has_intel_xpu_gpu() function.

Output:
The install detects Intel GPU (XPU) and installs torch 2.10.0+xpu correctly.

To confirm the llama.cpp backend is using SYCL:

source intel/oneapi/setvars.sh   # intel oneapi path
export LD_LIBRARY_PATH=~/.unsloth/llama.cpp/build/bin:$LD_LIBRARY_PATH   # the source build produces shared libs that need to be found
~/.unsloth/llama.cpp/build/bin/llama-server --list-devices

Output:

Available devices:
  SYCL0: Intel(R) Arc(TM) Pro B60 Graphics (24480 MiB, 24480 MiB free)
  SYCL1: Intel(R) Arc(TM) Pro B60 Graphics (24480 MiB, 24480 MiB free)
  SYCL2: Intel(R) Arc(TM) Pro B60 Graphics (24480 MiB, 24480 MiB free)
  SYCL3: Intel(R) Arc(TM) Pro B60 Graphics (24480 MiB, 24480 MiB free)