Refactorization of MapieRegressor and ConformityScore

[candicemyt]

Is your feature request related to a problem? Please describe.
The method predict of MapieRegressor compute the upper and lower bound of the prediction intervals not very intuitively. Further more, the ConformityScore class called by MapieRegressor does not allow to calculate the scores from X (only from y_pred). It is a problem to implement new conformity scores that need this feature (like Conformal Residual Fitting).

Describe the solution you'd like

• Refactorization of ConformityScore class : it would take X and estimator in parameters.
• Re-order the way conformity scores, quantile and bounds are calculated. It would be the conformity score, then the quantile and finally the bounds.