VAE Encode audio does not work reliable with the ComfyUI converted ace_1.5_vae.safetensors

### 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 VAE Encode should generate a latent which - back converted with VAE decode - should produce the original audio.

### Actual Behavior

In 10~20% the audio is empty (silent).

### Steps to Reproduce

See issue 12326, first post. This issue was unfortunately closed, even though the answers were off-topic and not addressing the bug from the OP.

### Debug Logs

```powershell
Nothing to see here.
```

### Other

* The bug is not related to the audio clip length. Short clips and long clips are affected as well.
* Cutting away 5 to 10 seconds of audio from the beginning or the end of the affected clip makes the encoding usually working, but that's not really a solution, because then these 5 to 10 seconds are missing.
* Whether tiled encoding is used or not also doesn't matter. Happens with and without tiling.