slowodown with nunchaku loader qwen image

### 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

it should not load qwen edit each time i generate image , it causes slodowns
If i do not touch any settings in qwen model loader - it should NOT CHECK or try to reload it, should load it from memory.
I tried disabled dynbamic vram in cmd but nothing changed.

### Actual Behavior

What happening with latest comfy? i have yuesterdays version and it is very slow loading qwen image edit compared to version from month ago and there is pointless debug print each time it generates, anyone havbe this issue recentlyu? is only option to downgrade from before the weird debugs ?
It basically reloads qwen image edit each time i generate the image which did not happen before and nodes were untouched .
RTX 3090 , 32 gbram .

### Steps to Reproduce

jusrt load qwen image edit with hunchaku qwen image loader and gguf clip loader

### Debug Logs

```powershell
model weight dtype torch.bfloat16, manual cast: None
model_type FLUX
🔍 Model wrapper type: 'NunchakuQwenImageTransformer2DModel'
🔍 Model wrapper module: M:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-nunchaku.models.qwenimage
🔍 Type repr:
🔍 Has 'model' attr? False
🔍 Has 'loras' attr? False
🔧 Wrapping NunchakuQwenImageTransformer2DModel with ComfyQwenImageWrapper
📦 Creating ComfyQwenImageWrapper with cpu_offload='disable'
🔧 CPU offload setting: 'disable' (VRAM margin: 4.0GB)
LoRA added: Qwen-Image-Edit-F2P.safetensors (strength=1.0)
🔍 Model wrapper type: 'ComfyQwenImageWrapper'
🔍 Model wrapper module: wrappers.qwenimage
🔍 Type repr:
🔍 Has 'model' attr? True
🔍 Has 'loras' attr? True
✅ Model is already wrapped (detected via attributes)
📦 Current CPU offload setting: 'disable'
LoRA added: Qwen-Image-Edit-F2P.safetensors (strength=0.65)
Requested to load NunchakuQwenImage
0%| | 0/5 [00:00 15GiB, disabling CPU offload

WARNING, you accessed scaled_fp8 from the model config object which doesn't exist. Please fix your code.

model weight dtype torch.bfloat16, manual cast: None
model_type FLUX
```

### Other

_No response_