pub fn normal_cdf(z: f64) -> f64
The standard normal CDF, through erfc(x) = Q(1/2, x²): accurate in the tails, where 1 - erf loses everything.
erfc(x) = Q(1/2, x²)
1 - erf