expm1x function

Accurate exp(x) - 1 - x (for smallish |x|)