pub(super) const REAL_ULPS: f64 = 64.0;
Rounding error of the real-space part of a 1D spherical sum relative to its moduli, in ulps: its Kambe integrals of even order keep their relative accuracy.