Offer late-integration models

[lgessl]

Offer models, i.e.,

1. a fitter function and
2. a method for the predict S3 generic from the stats package,

for late integration.

• The fitter takes a cross-validation object (typically zeroSumFit) and another fitter function (typically zeroSum::zeroSum()) and does another cross validation on the folds and their corresponding models from the cross-validation object. It returns another cross-validation object.
• For the second stage model, we shouldn't need regularization, i.e. no cross-validation. If this holds true, we needn't be concerned about optimization.

Doing this for zeroSum has the highest priority.

[lgessl]

nested_pseudo_cv() does this in a more general setting. With the help of ptk_zerosum and ptk_ranger, you can easily nested a logistic model into another logistic model or a random forest. More about this in the release notes of verision 0.5.0.