pub fn normal_cdf(x: f64) -> f64
Standard normal CDF Φ(x) using rational approximation.
Accuracy: |error| < 7.5e-8 for all x (Abramowitz & Stegun 26.2.17).