Latent From Batch does not reproduce original image from batch

### 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

When generating a batch of images, the seed is shared with all of the images.

In order to reproduce a specific image in the batch, you must use the `Latent From Batch` node.

### Actual Behavior

However, the `Latent From Batch` node does not actually reproduce the image from the original batch.

### Steps to Reproduce

Run this workflow image (this will generate a batch of 4 images):

Then run this workflow image (this will generate a single image):

The single image is similar to the batched image, but they are not identical.

In some cases the single image can be *significantly* different from the batched image.

### Debug Logs

```powershell
** Platform: Linux
** Python version: 3.10.16 | packaged by conda-forge | (main, Dec 5 2024, 14:16:10) [GCC 13.3.0]
** Python executable: /home/pauan/.config/pinokio/api/comfy.git/app/env/bin/python
** ComfyUI Path: /home/pauan/.config/pinokio/api/comfy.git/app
** ComfyUI Base Folder Path: /home/pauan/.config/pinokio/api/comfy.git/app
** User directory: /home/pauan/.config/pinokio/api/comfy.git/app/user
** ComfyUI-Manager config path: /home/pauan/.config/pinokio/api/comfy.git/app/user/default/ComfyUI-Manager/config.ini
** Log path: /home/pauan/.config/pinokio/api/comfy.git/app/user/comfyui.log

Checkpoint files will always be loaded safely.
Total VRAM 15838 MB, total RAM 96175 MB
pytorch version: 2.7.0+cu128
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 5070 Ti : cudaMallocAsync
Using async weight offloading with 2 streams
Enabled pinned memory 91365.0
Using pytorch attention
Python version: 3.10.16 | packaged by conda-forge | (main, Dec 5 2024, 14:16:10) [GCC 13.3.0]
ComfyUI version: 0.3.76
ComfyUI frontend version: 1.33.10
[Prompt Server] web root: /home/pauan/.config/pinokio/api/comfy.git/app/env/lib/python3.10/site-packages/comfyui_frontend_package/static
### ComfyUI Version: v0.3.76-4-ga17cf1c3 | Released on '2025-12-01'

Context impl SQLiteImpl.
Will assume non-transactional DDL.
No target revision found.
```

### Other

I have tried different models, samplers, schedulers, and batch sizes. They all reproduce this bug.