Investigate better way to make integer grid values for partial dependence

[tamargrey]

With #4095, we made the handling of integer grid values for partial dependence explicit, rounding the fractional values produced in _grid_from_X to allow IntegerNullable types to be used for partial dependence.

One downside of this method for determining grid values is that it's possible for us to end up with duplicate grid values if, say, two grid values were calculated to be 4.1 and 4.4 and then both got rounded down to 4. We should investigate if there is a better way to get integer grid values for partial dependence calculations or if there is a way for us to maintain the fractional grid values and pass those into our pipelines (this is currently not possible because of our _schema_is_equal check when transforming on new data).