Add implementation of ChemDisGene data set

[mariosaenger]

Closes #917

[leonweber]

Hi, thanks for your contribution! It seems like the dataset statistics reported by the unit tests do not match those reported in the source paper (https://aclanthology.org/2022.lrec-1.116.pdf):

train
==========
id: 523
document_id: 523
passages: 1046
entities: 14248
normalized: 13686
events: 0
coreferences: 0
relations: 59414

Is this expected? Am I missing something?

[mariosaenger]

Thanks for checking the implementation. There are several aspects to keep in mind. First the dataset consists of a curated and a non-curated part. This implementation only concerns the former one. Second, the data set annotates relations only on abstract-level (using knowledge base identifiers). Following default practices in BigBio, I unrolled the document-level relations to mention-level. Note, however, the document-level annotations are available in the source schema. These aspects complicate a direct comparison of the numbers :-/

[leonweber]

Ah, thanks for pointing this out. Then let's merge this : )

[leonweber]

LGTM