Modify our code to initialize woodwork with partial schemas

[freddyaboulton]

Woodwork introduced the ability to initialize typing information with a partial schema in alteryx/woodwork#1100

This means that we can potentially get rid of _retain_custom_types_and_intialize_woodwork and therefore retain more than just the logical types after some components modify the input data in a way that breaks the woodwork schema.

For example, this is how the TargetEncoder would look:

    X_t_df = pd.DataFrame(X_t, columns=X_ww.columns, index=X_ww.index)
    cat_columns = X_ww.ww.select("categorical", return_schema=True).columns
    columns = [c for c in X_t_df.columns if c not in cat_columns]
    # Infer types on newly created columns
    X_t_df.ww.init(schema=original_schema._get_subset_schema(columns))

[chukarsten]

As acceptance criteria for completing this issue, let's evaluate the ability to remove _retain_custom_types_and_intialize_woodwork and refactor using the new WW capabilities.

[chukarsten]

@freddyaboulton do we need to look at the fix in #2752 to refactor to use this as well?

[freddyaboulton]

@chukarsten If you're asking if we're going to have to revisit the Imputer as part of this refactor, then you're spot on. I think anywhere _retain_custom_types_and_initalize_woodwork is in scope for this issue!