better example for crossvalidattion + other ideas

• more detailed explanation : fit happens inside function call for cross_val score
• digits classification in examples

KNN: diagram with one circle scaled.

linear classifiers: coefficients have meaning (weights).

logreg: regress on class probabilities + threshold.

classifier: full optional sections with theory

rbf: iso maps, also for loss + penalty landscape.

classifiers: first general idea with plots, then parameters of the method, then theory.

todo: pydot into conda env yaml file