BPE the identifier dataset and publish the new model

[vmarkovtsev]

As discussed on the reading club, we should run https://github.com/google/sentencepiece on our identifiers dataset and produce a nice compact ASDF in Modelforge to embed any identifier.
That model can be further integrated somewhere near our ID splitter.

[vmarkovtsev]

Assigning myself - please let me do something interesting along with the endless meetings, hehe.