Add underscores at the end of fitted attributes so scikit-learn can determine fit status

[lukethomrichardson]

Hello, thanks for maintaining a great package. Scikit-learn determines if an estimator is fitted by evaluating whether it contains attributes that end with an underscore (e.g., self.alpha_). Currently, absence of these attributes lead scikit-learn to conclude that pipelines that end with a WrapClassifier (or other crepes predictor object) are not fitted, even if the wrapped learner is fitted. I recommend including these underscored attributes in variables that are fitted in the calibrate methods of WrapClassifier (and others where appropriate). According to sklearn FutureWarnings in 1.6.1, this will start to cause problems in sklearn 1.8.0.