When using the Model MergeFlux1 node I don't think weight_dtype is applied properly.

### Expected Behavior

Set Weight_dtype to Default and attempt to merge,
The 32G model was output, and quantization was possible.

### Actual Behavior

We tried to merge by setting Weight_dtype to Default,
The 20GB model is output, and quantization is not possible.

### Steps to Reproduce

1. Register 11G model in slot 1
2. Register 32G model in slot 2
3. Set Weight_dtype to Default
4. Select and merge a high-capacity clip model
5. The printed model should be 32G.

### Debug Logs

```powershell
No special debug logs.
```

### Other

Before the update, if you have a 32G model regardless of the slot,
All of them have 32G models, and they're quantizable,

after the update
If you put 11G in slot 1, and 32G model in slot 2,
The 20G model is created, and it becomes non-quantizable.

However, on the contrary, if you put the 32G model in slot 1, and the 11G model in slot 2,
The 32G model is created, and it becomes quantizable.

Please check if the bug has occurred or if it has been fixed.