Minor open questions and ideas

[erikcs]

User facing

• cv.balnet: path error plot() instead? (update surrounding text in manuscript if so)
• cv.balnet: maybe make min.error a public attribute (could for ex use it to tune alpha)
• cv.balnet could save fold predictions for downstream cross-fitting use cases
• Maybe optional test set W arg to predict contrast weights/propensities

Misc. internal

• summary coef/balweights contrast: could have returned list
• save W.hat instead of X.orig reference?
• could save stan w/o X and use later on in diagnostics
• Maybe have separate files for print/plot etc
• Add test for group options, penalty works as expected
• Maybe strip away glms/solvers from adelie_core that are not needed

Misc. doc

• doc: Predict docstring with w=0/1 distinction + test set example?
• doc: note that if standardize=F, metrics like "SMD" raw imbalance (eg overloaded term)