Generation grinds to a halt when typing text or if focus is inside of the clip node

### Expected Behavior

Just be able to edit clip while it generates an image.

### Actual Behavior

If I type while it generates the image, each step will take forever (20 seconds to a few minutes) to generate. If I click outside on the background before generating, each steps will take exactly 3 seconds.

### Steps to Reproduce

[workflow.json](https://github.com/user-attachments/files/16691498/workflow.json)

### Debug Logs
While typing:
```powershell
got prompt
loaded partially 20529.491 20520.070434570312 0
100%|██████████████████████████████████████████████████████████████████████████████████| 16/16 [03:15<00:00, 12.23s/it]
Requested to load AutoencodingEngine
Loading 1 new model
Prompt executed in 204.79 seconds
```
When not typing with focus outside.
```powershell
got prompt
Requested to load FluxClipModel_
Loading 1 new model
loaded partially 20529.491 20520.070434570312 0
100%|██████████████████████████████████████████████████████████████████████████████████| 16/16 [00:41<00:00, 2.60s/it]
Requested to load AutoencodingEngine
Loading 1 new model
Prompt executed in 55.38 seconds
```

### Other

_No response_