I went back and forth on a lot of things here but ultimately I think I want us to move over to FastAPI before I mess with the architecture any more. On the bright side, I think this resolves #29.

• Currently just accessing the normalizer via REST. I don't know if it makes sense to try to run them both together in one Python process. It sort of depends on how much Dockerizing we want to do and how we want to do it. I'll keep Modularize translator/data proxy layer #24 open for now.
• I don't know how essential the sequence stuff is, but it unfortunately requires a separate seqrepo dataproxy because the variation normalizer doesn't quite expose the same stuff under translate_identifier. It'd be nice to just drop it so that we have one less thing opening seqrepo files.