Save Checkpoint and ModelSave nodes crash ComfyUI when saving Flux Dev bfloat16 + Lora

### Expected Behavior

Save Checkpoint and ModelSave nodes should not crash ComfyUI when saving Flux Dev bfloat16 + Lora

### Actual Behavior

Save Checkpoint and ModelSave nodes crash ComfyUI when saving Flux Dev bfloat16 + Lora.
I noticed that VRAM consumption keeps growing until it reaches the 16GB limit, and then it crashes.
I tested with different Loras, for example: [Example Lora](https://huggingface.co/alexds9/Flux_Lora_Test_01/blob/main/model_15.safetensors)
Simple workflow:
[flux+lora_modelsave_simple_workflow.json](https://github.com/user-attachments/files/16674948/flux%2Blora_modelsave_simple_workflow.json)
I tested it with all custom nodes disabled, but it still crashed.
But when I tested it without loading a Lora, it was able to save the model, so it seems that loading the Lora is contributing to the failure.

### Steps to Reproduce

Loading Flux Dev bfloat16 + Lora and saving it with Save Checkpoint and ModelSave.
Workflow and info attached.

### Debug Logs

```powershell
got prompt
model weight dtype torch.bfloat16, manual cast: None
model_type FLUX
Requested to load FluxClipModel_
Loading 1 new model
loaded completely 0.0 4777.53759765625 True
clip missing: ['text_projection.weight']
Requested to load Flux
Loading 1 new model
loaded partially 14695.825 14688.193481445312 0
Killed
```

### Other

_No response_