add tokenizer rules for normalizing semantic variants

[thatbudakguy]

using "semantic variant" data from Unihan, we can maybe start to fix problems of differing glyphs at the tokenizer level – see the NewNLP docs and probably also spaCy's docs on tokenizer exceptions. We'd basically want to add NORM forms for each semantic variant.

this wouldn't change token.text, however, so we might need to figure out the best way to use it.