Update EvalML to be compatible with the new Woodwork `Boolean` inference

[ParthivNaresh]

Woodwork updated their Boolean inference in this PR, with an additional update in this one.

This issue covers two main areas of change:

• Updating tests to be compatible
  • Replacing instances of y.ww.init() with ww.init_series(y)
  • Replacing expected int64 with bool
  • Full list here
• Updating component behaviour
  • CatBoostRegressor - Target value "False" cannot be parsed as float in fit
  • LightGBMClassifier and XGBoostClassifier - Possibly an issue with LabelEncoder in predict
  • Oversampler - SMOTENC returned instead of SMOTE due to difference in expected categorical columns
  • ClassImbalanceDataCheck - Reverse mapping required to return the original values instead of True/False in data check details/message
  • TargetLeakageDataCheck - Issue when calculating dependence when features or target are inferred as boolean instead of int
  • partial_dependence - Inconsistent logical types when attempting pred = prediction_method(X_eval)
  • TimeSeriesPipelineBase - Attempting to use the old schema when reinitializing y can lead to an incomaptible dtype error in _drop_time_index
  • ClassificationPipeline - Mapping issue in LabelEncoder that results in incorrect mapping when _encode_targets is called. The keys in the mapping dict can't be found so post mapping all values are converted to NaN