LTXVImgToVideoInplace produces black latent

### 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 Node `LTXVImgToVideoInplace` should generate a latent filled with the input image.

### Actual Behavior

The node produces only black output, with possible mono colored frames at the end.

### Steps to Reproduce

Use the template workflow for LTXV2.5 image-to-video on Windows with an AMD GPU (in my case gfx1100).

### Debug Logs

```powershell
No error messages or specific warnings:

[INFO] Prompt executed in 1.20 seconds
[INFO] got prompt
[INFO] Requested to load CausalDiffusionVAE
[INFO] loaded completely; 1403.92 MB loaded, full load: True
[INFO] Requested to load LTXAV
[INFO] loaded completely; 17977.32 MB usable, 15068.95 MB loaded, full load: True
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:55<00:00, 6.98s/it]
[INFO] Requested to load LatentUpsampler
[INFO] Model LatentUpsampler prepared for dynamic VRAM loading. 949MB Staged. 0 patches attached. Force pre-loaded 34 weights: 68 KB.
[INFO] Requested to load CausalDiffusionVAE
[INFO] loaded completely; 1403.92 MB loaded, full load: True
[INFO] Requested to load LTXAV
[INFO] loaded completely; 16967.27 MB usable, 15068.95 MB loaded, full load: True
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:33<00:00, 11.17s/it]
[INFO] Requested to load CausalDiffusionVAE
[INFO] loaded completely; 1403.92 MB loaded, full load: True
[INFO] Requested to load AudioVAE
[INFO] loaded completely; 346.73 MB loaded, full load: True
[INFO] Prompt executed in 141.35 seconds
```

### Other

Text-to-Image works fine, also using different Diffusion Models did not help.

Currently tried on Windows 11, with an AMD 7900XT. At least from build `783545f` onwards, although I am sure that this also applied to earlier builds.

I am not sure if it is a general problem, but Minimax H3 also produced a black video.

Start with the LTXVImgToVideoInplace node and reproduce the LTXV2.5 image-to-video template workflow on Windows with an AMD 7900XT. Compare the generated latent and decoded frames with the input image; the issue is done when the latent is no longer black and the resulting video has usable frames.

Written by the indexing model from the issue text.