Add model training & prediction endpoints to the library

[yonromai]

TL;DR:

from nxontology_ml.data import get_efo_otar_slim
from nxontology_ml.model.predict import train_predict

nxo = get_efo_otar_slim()
df = train_predict(nxo=nxo)

(From readme)

Some notes:

• More context can be found in this comment
• The output of the prediction should be consistent when invoked from the same system (in new python session). From experience, it's likely to differ slightly across platform
• No file is written (responsibility of the caller), but the content of the data is identical to data/efo_otar_slim_v3.57.0_precisions.tsv
• FWIW, schemaless exports make me a little nervous - but I guess they seem to be the norm in the data-science world, so it's on me

[dhimmel]

FWIW, schemaless exports make me a little nervous - but I guess they seem to be the norm in the data-science world

I do not disagree. We often use pydantic internally, although the ability to apply it recordwise to a pandas dataframe is somewhat lacking. On the other hand, if this pipeline breaks, it's not a big deal as long as that results in failure and not bad output in a silent manner.