Extract and Save Lora is broken

### 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

The extracted lora works when applied to the model.

### Actual Behavior

Applying the extracted lora to the model produces noise.

### Steps to Reproduce

Use a recent version of ComfyUI to extract a lora for the difference between Anima-turbo and Anima-base. Apply the lora to Anima-base and run inference. The output will be noise.
Use an older version of ComfyUI (I used v0.22.0) and similarly extract a lora and run inference. The lora will work as expected.

### Debug Logs

```powershell
[INFO] pytorch version: 2.13.0+cu132
[INFO] Set vram state to: NORMAL_VRAM
[INFO] Device: cuda:0 NVIDIA GeForce RTX 3090 : cudaMallocAsync
[INFO] Device: cuda:1 NVIDIA GeForce RTX 5060 Ti : cudaMallocAsync
[INFO] Using async weight offloading with 2 streams
[INFO] Using pytorch attention
[INFO] Using Comfy Kitchen attention
[INFO] aimdo: src-win/cuda-detour.c:38:INFO:aimdo_setup_hooks: installing 6 hooks
[INFO] aimdo: src/control.c:262:INFO:comfy-aimdo NVML pressure enabled
[INFO] aimdo: src-win/shmem-detect.c:80:INFO:comfy-aimdo WDDM adapter match: NVIDIA GeForce RTX 3090 runtime_luid=00000000:0000fd1b dxgi_luid=00000000:0000fd1b
[INFO] aimdo: src/control.c:277:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 3090 (VRAM: 24575 MB)
[INFO] aimdo: src/control.c:262:INFO:comfy-aimdo NVML pressure enabled
[INFO] aimdo: src-win/shmem-detect.c:80:INFO:comfy-aimdo WDDM adapter match: NVIDIA GeForce RTX 5060 Ti runtime_luid=00000000:000109c7 dxgi_luid=00000000:000109c7
[INFO] aimdo: src/control.c:277:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 5060 Ti (VRAM: 16310 MB)
[INFO] DynamicVRAM support detected and enabled
[INFO] Python version: 3.14.7 (tags/v3.14.7:823f032, Aug 5 2026, 10:51:32) [MSC v.1944 64 bit (AMD64)]
[INFO] ComfyUI version: 0.33.0
[INFO] comfy-aimdo version: 0.4.13
[INFO] comfy-kitchen version: 0.2.31
[INFO] comfyui-frontend-package version: 1.49.6
[INFO] comfyui-workflow-templates version: 0.11.44
[INFO] comfyui-embedded-docs version: 0.5.10
[INFO] comfy-kitchen version: 0.2.31
```

### Other

The v0.22.0 version I used was using a different venv with pytorch version: 2.13.0+cu130. The difference in venv is not the cause, as the same issue was reproduced on this venv as well.
Multiple GPUs are not the cause either, as I tried setting it to use only 1 GPU with `--cuda-device 0` and the issue persisted.

Start by reproducing the reported workflow: extract a LoRA for the difference between Anima-turbo and Anima-base in a recent ComfyUI version, then apply it to Anima-base and run inference. Compare the result with the same workflow in v0.22.0; done means the extracted LoRA produces a usable output rather than noise.

Written by the indexing model from the issue text.