Optimize csvcut -n by bypassing type-inference

Created by: JoeGermuska

I kind of think csvcut -n should be fast. If inference were not default, you could stop after reading the first row. What do you think to making it only on request, or barring that, having an option that suppresses it?