"[I]f we make the reasonable ballpark assumption that people working in a given field tend to come from the 25% of people most interested in that field, sex differences in occupational interests would account for the entirety of the engineering gender gap and much of the gap in science and mathematics."