Python Pandas with 1 milion rows. Add another calculated column faster. Is faster by map or apply?