Remove BooleanNullable Logical Type and have 1 Boolean Logical Type

- We originally had **BooleanNullable** and **Boolean** because EvalML relies on estimator/classifiers that do not support the pandas `boolean` dtype (which is used in **BooleanNullable**)
- However, a better long term solution is for Woodwork to have 1 Boolean type that supports nullable values
- For this issue:
1. Remove/Deprecate BooleanNullable
2. Change the Boolean physical type from `int64` to `Int64`
3. Update inference logic to account for this change
4. Verify that Featuretools and EvalML can properly handle this change