[ENH] Break down logic of optimization and experiments in optimizers

[ArtemLiA]

Hi! Currently logging and processing of all experimental results is carried out inside optimizers. It looks suboptimal, because, for example, it is limits and evaluating and logging of new metrics (we need rewrite the logic of optimizer). I propose to separate these two tasks and propose two main variants:

Option 1

• Each optimizer has a method step(*args, **kwargs), which performs an optimization step and each optimizer has method/attribute/property get_best_params (or other name), which returns best parameters set and objective function value at this params point.
• We have a separate class Experiment, which performs many optimization steps, logs and evaluates the results.

Option 2

• The optimize method of each optimizer has a parameter callback, which is callback-function. it logs optimization results after each step and stop futher optimization by returning True value. This logic is used in scipy.optimize package.