Windows + CUDA 13: SageAttention import crashes (DLL load failed: _fused) after ComfyUI update

### Custom Node Testing

- [x] I have tried disabling custom nodes and the issue persists (see [how to disable custom nodes](https://docs.comfy.org/troubleshooting/custom-node-issues#step-1%3A-test-with-all-custom-nodes-disabled) if you need help)

### Expected Behavior

Sage attention works

### Actual Behavior

On Windows, after updating ComfyUI to v0.8.2 (torch 2.9.1 + cu130), ComfyUI fails to start because diffusers eagerly imports sageattention. The import crashes with DLL load failed while importing _fused, as there is currently no compatible Windows binary for SageAttention with torch 2.9 / CUDA 13 (RTX 50-series).

2026-01-08 22:48:19.553] bitsandbytes library load error: Configured CUDA binary not found at C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\bitsandbytes\libbitsandbytes_cuda130.dll
If you are using Intel CPU/XPU, please install intel_extension_for_pytorch to enable required ops
[2026-01-08 22:48:20.466] Traceback (most recent call last):
File "C:\Users\user\AppData\Local\Programs\ComfyUI\resources\ComfyUI\nodes.py", line 2155, in load_custom_node
module_spec.loader.exec_module(module)
File "", line 999, in exec_module
File "", line 488, in _call_with_frames_removed
File "C:\Users\user\Documents\ComfyUI\custom_nodes\ace-step\__init__.py", line 1, in
from .ace_step_nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
File "C:\Users\user\Documents\ComfyUI\custom_nodes\ace-step\ace_step_nodes.py", line 17, in
from ace_step.pipeline_ace_step import ACEStepPipeline as AP
File "C:\Users\user\Documents\ComfyUI\custom_nodes\ace-step\ace_step\pipeline_ace_step.py", line 9, in
from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3 import retrieve_timesteps
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\pipelines\stable_diffusion_3\pipeline_stable_diffusion_3.py", line 31, in
from ...models.autoencoders import AutoencoderKL
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\autoencoders\__init__.py", line 1, in
from .autoencoder_asym_kl import AsymmetricAutoencoderKL
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\autoencoders\autoencoder_asym_kl.py", line 23, in
from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution, Encoder, MaskConditionDecoder
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\autoencoders\vae.py", line 25, in
from ..unets.unet_2d_blocks import (
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\unets\__init__.py", line 6, in
from .unet_2d import UNet2DModel
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\unets\unet_2d.py", line 24, in
from .unet_2d_blocks import UNetMidBlock2D, get_down_block, get_up_block
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\unets\unet_2d_blocks.py", line 36, in
from ..transformers.dual_transformer_2d import DualTransformer2DModel
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\transformers\__init__.py", line 20, in
from .transformer_bria import BriaTransformer2DModel
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\transformers\transformer_bria.py", line 14, in
from ..attention_dispatch import dispatch_attention_fn
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\attention_dispatch.py", line 92, in
from sageattention import (
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\sageattention\__init__.py", line 1, in
from .core import sageattn, sageattn_varlen
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\sageattention\core.py", line 47, in
from .quant import per_block_int8 as per_block_int8_cuda
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\sageattention\quant.py", line 20, in
from . import _fused
ImportError: DLL load failed while importing _fused: The specified module could not be found.

### Steps to Reproduce

On launching comfyui desktop

### Debug Logs

```powershell
2026-01-08 22:48:19.553] bitsandbytes library load error: Configured CUDA binary not found at C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\bitsandbytes\libbitsandbytes_cuda130.dll
If you are using Intel CPU/XPU, please install intel_extension_for_pytorch to enable required ops
[2026-01-08 22:48:20.466] Traceback (most recent call last):
File "C:\Users\user\AppData\Local\Programs\ComfyUI\resources\ComfyUI\nodes.py", line 2155, in load_custom_node
module_spec.loader.exec_module(module)
File "", line 999, in exec_module
File "", line 488, in _call_with_frames_removed
File "C:\Users\user\Documents\ComfyUI\custom_nodes\ace-step\__init__.py", line 1, in
from .ace_step_nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
File "C:\Users\user\Documents\ComfyUI\custom_nodes\ace-step\ace_step_nodes.py", line 17, in
from ace_step.pipeline_ace_step import ACEStepPipeline as AP
File "C:\Users\user\Documents\ComfyUI\custom_nodes\ace-step\ace_step\pipeline_ace_step.py", line 9, in
from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3 import retrieve_timesteps
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\pipelines\stable_diffusion_3\pipeline_stable_diffusion_3.py", line 31, in
from ...models.autoencoders import AutoencoderKL
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\autoencoders\__init__.py", line 1, in
from .autoencoder_asym_kl import AsymmetricAutoencoderKL
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\autoencoders\autoencoder_asym_kl.py", line 23, in
from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution, Encoder, MaskConditionDecoder
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\autoencoders\vae.py", line 25, in
from ..unets.unet_2d_blocks import (
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\unets\__init__.py", line 6, in
from .unet_2d import UNet2DModel
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\unets\unet_2d.py", line 24, in
from .unet_2d_blocks import UNetMidBlock2D, get_down_block, get_up_block
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\unets\unet_2d_blocks.py", line 36, in
from ..transformers.dual_transformer_2d import DualTransformer2DModel
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\transformers\__init__.py", line 20, in
from .transformer_bria import BriaTransformer2DModel
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\transformers\transformer_bria.py", line 14, in
from ..attention_dispatch import dispatch_attention_fn
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\diffusers\models\attention_dispatch.py", line 92, in
from sageattention import (
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\sageattention\__init__.py", line 1, in
from .core import sageattn, sageattn_varlen
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\sageattention\core.py", line 47, in
from .quant import per_block_int8 as per_block_int8_cuda
File "C:\Users\user\Documents\ComfyUI\.venv\Lib\site-packages\sageattention\quant.py", line 20, in
from . import _fused
ImportError: DLL load failed while importing _fused: The specified module could not be found.
```

### Other

_No response_