Likewise for greater-or equal?*. The reverse refinement on compare is only used for not-equal?* which seems a waste.
Actually, I think it would be better to let pass all the COMP_* actions to the datatype and let each datatype decide how to process them optimally.
FIX: added COMP_NOT_EQUAL as standalone comparison mode (related to i…