Add XPU device support for PD disaggregation with NIXL backend

[Copilot]

NIXL backend for PD disaggregation was hardcoded to CUDA. This adds Intel XPU support.

Changes

• Device detection (transfer_utils.py): Extended get_correct_device() to handle xpu device strings
• NIXL transfer channel (nixl_channel.py):
  • Added device parameter to NixlChannel and NixlAgentWrapper
  • Dynamic mem_type determination: cuda/xpu/cpu based on device string
  • Validates device types, raises on unsupported values
• Device initialization: Updated all NIXL code paths to support XPU
  • pd_backend.py: Passes device to transfer channel, uses torch.xpu.set_device() when applicable
  • p2p_backend.py: Explicitly passes device="cpu" for CPU-based transfers
  • nixl_storage_backend.py: XPU device setting support
  • cache_engine.py: XPU support in NIXL allocator
• Error handling: Checks hasattr(torch, "xpu") before XPU operations with clear error messages

Usage

pd_buffer_device: "xpu"  # Previously only "cuda" or "cpu"
transfer_channel: "nixl"
nixl_backends: ["UCX"]

Backward compatible: defaults to CUDA when device not specified.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[github-actions]

This pull request has been automatically marked as stale because it has not had activity within 60 days. It will be automatically closed if no further activity occurs within 30 days.