Description
In sklearn BaseSearchCV.fit, which calls _validation._fit_and_score() with a cloned estimator for each parallel call, the cloned estimators have a new instance of column_sharer created for each subobject rather than a single instance shared by all of them (the different steps in datapreparer now do not share similar instances, so it is not a shared instance object anymore). This creates issues as when we reset a new intent in the column_sharer, it does not affect the scope or say, preprocessor.