Draft linear model residual splitting

[mnwright]

Quick and dirty implementation of splitting on residuals of linear models in the nodes. Usage examples:

ranger(Sepal.Length ~ ., iris, confounders = iris[, 5, drop = FALSE])
ranger(Sepal.Length ~ ., iris, confounders = iris[, 4:5], splitrule = "maxstat", alpha = 0.05)

[mnwright]

Now the variable importance and prediction also uses the linear models:

rf <- ranger(Sepal.Length ~ ., iris, confounders = iris[, 5, drop = FALSE], importance = "permutation")
rf$variable.importance

To predict, also give the confounders to predict():

predict(rf, iris, confounders =  iris[, 5, drop = FALSE])