My Environment class is starting to be too messy, especially regarding how I handle the different cases of stationary and non stationary bandits.

• I should remove this optimization of self._historyOfChangePoints and self._historyOfMeans dictionaries in NonStationaryMAB, and just store the full list of size K * T of means!
• I should just store this full list (don't care about size, it will anyway be of the order of T which is the storage space needed to store the actual rewards!)
• And trick the .maxArm attribute of env in Evaluator so that the previously written methods for regret / last regret / accumulated rewards etc are the same.

For plots

• Giving number of break points in the ylabel is fine, but maybe I could add red star markers on the xaxis at each breakpoint ? Or dashed vertical line?
• Fixed for regret (I think, at least for PieceWiseStationaryMAB)
• TODO fix for best arm pull selections!