handle character variants when loading data

[thatbudakguy]

see https://github.com/direct-phonology/core/blob/6a800a3201de43c039a6f7f096aef3a65a843922/core/bin/gentable.py#L84-L134

[thatbudakguy]

see also spacy's own docs on data augmentation which lets you swap variants to get a more robust training process. our existing variant file could be converted into this format pretty easily, and then we could use the orth_variants augmenter or define our own.