fix(group_offloading): use Ascend NPU stream when available

`apply_group_offloading` with `use_stream=True` only created a stream
for CUDA or Intel XPU devices, and raised
"Using streams for data transfer requires a CUDA device, or an Intel
XPU device." on every other accelerator. On Ascend NPU, `torch.npu.Stream()`
is available (torch_npu exposes it), so stream-based onload/offload should
be supported there too.

Adds an `npu` branch consistent with the existing CUDA/XPU checks, and
updates the error message to mention the Ascend NPU device.

Author: li-lizhe

src/diffusers/hooks/group_offloading.py

@@ -670,8 +670,12 @@ def apply_group_offloading(
             stream = torch.cuda.Stream()
         elif hasattr(torch, "xpu") and torch.xpu.is_available():
             stream = torch.Stream()
+        elif hasattr(torch, "npu") and torch.npu.is_available():
+            stream = torch.npu.Stream()
         else:
-            raise ValueError("Using streams for data transfer requires a CUDA device, or an Intel XPU device.")
+            raise ValueError(
+                "Using streams for data transfer requires a CUDA device, an Intel XPU device, or an Ascend NPU device."
+            )
 
     if not use_stream and record_stream:
         raise ValueError("`record_stream` cannot be True when `use_stream=False`.")