What are the assumptions for the distribution of the features for regression models like Lasso regression or Ridge regression? Why is it better to have features with Gaussian distributions?