Autofeat

- changed autofeat model to differentiate between regression and classification tasks, adding the `AutoFeatRegressor` and `AutoFeatClassifier` classes
- simplified feature selection process

- more robust featsel with noise filtering

- change default value for `feateng_steps` to 2, in line with results on realworld datasets

- make feature selection less prone to overfitting

- add `FeatureSelector` class to use feature selection separately
- make feature selection more robust and move into featsel
- make the models more sklearn like and test with sklearn estimator tests
- replace sympy's ufuncify with lambdify
- better logs
- use immutable default arguments
- make pi theorem optional
- handle nans in transform

- updated documentation