[helion] Triton API change for min_dot_size

[Stonepia]

🐛 Describe the bug

Triton recently changed their API in intel/intel-xpu-backend-for-triton#5329

This PR moved min_dot_size() into class XPUBackend. This made helion call failed in:

 from triton.backends.intel.compiler import min_dot_size as min_dot_size_xpu

https://github.com/pytorch/helion/blob/b9925d9dd5c1658e532f0b667604bfe41c474eb4/helion/_compat.py#L256

We need to change to

from triton.backends.intel.compiler import XPUBackend

current_device = torch.xpu.current_device()

xpu_backend = XPUBackend(current_device)
min_dot_size_xpu = xpu_backend.min_dot_size

We may need an API from pytorch side to allow device_properties be passed elegantly.

Versions

torch: 2.10.0.dev20251123+xpu
triton : pytorch-triton-xpu 3.5.0+git1b0418a9
helion: b9925d9dd5c1658e532f0b667604bfe41c474eb4

[Stonepia]

The change is in:
pytorch/helion#1173

[Stonepia]

The triton issue is in:
intel/intel-xpu-backend-for-triton#5574