make `betaboost(y = y, x = x) ` possible. Needs to differentiate between fitting mu and phi, and linear and non-linear base-learners.