feat: add chebi 20 to datasets

[jackapbutler]

• closes New Task: Add chebi-20 dataset #63

I've added a transform.py script and related meta.yaml for the Chebi-20 dataset. It contains a dataset of SMILES with their corresponding natural language description. The dataset was first used in this paper but also used here by the same authors. It also features in Kevin's awesome-chemistry-datasets repository.

I reopened this PR because I accidentally removed the commits from the old one 🙃.

[kjappelbaum]

Thanks, looks quite nice and clean to me! The only thing is the train/val/test thing. I have no strong preference other than that it should be consistent.

In https://github.com/OpenBioML/chemnlp/pull/98/files# we have been introducing a special column indicating the split. We might also try your approach and revert the other PR, do you have any preference or opinion?

[jackapbutler]

Thanks, looks quite nice and clean to me! The only thing is the train/val/test thing. I have no strong preference other than that it should be consistent.

In https://github.com/OpenBioML/chemnlp/pull/98/files# we have been introducing a special column indicating the split. We might also try your approach and revert the other PR, do you have any preference or opinion?

I've replies in #104 but I think they're two different things

[kjappelbaum]

Looks good to me, thanks! 💯