Currently the nil values in DataFrame are UndefinedObject instance. We would like to replace them by something like DataNaN that could do computations.

For example:

NaN - 5 => NaN

This would remove the need to nil check everywhere and simplify the code