Is class numbering with gaps supported for training on custom datasets?

### Checklist

- [X] I have searched for [similar issues](https://github.com/isl-org/Open3D-ML/issues).
- [X] I have tested with the [latest development wheel](http://www.open3d.org/docs/latest/getting_started.html#development-version-pip).
- [X] I have checked the [release documentation](http://www.open3d.org/docs/release/) and the [latest documentation](http://www.open3d.org/docs/latest/) (for `main` branch).

### My Question

Hi,

I have data following the standard Las classification, meaning that some indices are unused or reserved. If possible I would like to use this data directly and only configure dataset and pipeline to handle the class to index mapping.
I am fine-tuning a KP-Conv model with the output layer modified to fit our number of classes.

From the CustomDataset3D [label_to_index](https://github.com/isl-org/Open3D-ML/blob/b64b51445bf2fdb17f7de4fd7a1c383ac238a3dc/ml3d/datasets/customdataset.py#L115) - template I assume that a mapping from a arbitrary class number to the consecutive output-channel index is somehow supported.

When running the training pipeline, I get an index out of bounds error [here](https://github.com/isl-org/Open3D-ML/blob/b64b51445bf2fdb17f7de4fd7a1c383ac238a3dc/ml3d/torch/modules/losses/semseg_loss.py#L35), since the 'label' == 65 is greater than my number of classes.

It is not a big deal to write the processing to re-index the class numbers. I am just wondering: is this functionality is actually supported? And, I am missing something completely?

Thanks in advance for any support