pub fn max_f64(a: f64, b: f64) -> f64
Computes the maximum of two f64 values, being careful about NaNs.