sinc_pi can be more precise with Horner's method

[minorlogic@…]

double sinc1 = 1.0 + x2 * (-1.0 + x2 / 20.0) / 6.0;

double sinc2 = 1.0 - x2 /6.0 + x2*x2 / 120.0;

in some cases sinc1 provides better precision compared to sinc2

x generated with

double x = sqrt(sqrt(std::numeric_limits<double>::epsilon())); for (int i = 0; x > sqrt(std::numeric_limits<double>::epsilon()); i++) {

x *= 0.99876765378657564;

}

[John Maddock]

Thanks, applied in ​https://github.com/boostorg/math/commit/658945d50822adae83d77083a559e785ce18609c along with better tests.