Extend support for float/double tolerances in testing