output of DiffUnet.forward_diffusion() has 6 channels instead of 3

The output of

```
import deepinv
import torch

denoiser = deepinv.models.DiffUNet(large_model=False)
denoiser.eval()

x = torch.randn(1, 3, 128, 128)
denoiser.forward_diffusion(x, torch.tensor([0.1])).shape
```

is `torch.Size([1, 6, 128, 128])` (6 channels instead of 3).

Any idea where that comes from?


cc @annegnx