Optimize conversion of float to f64 (and PyFloat::value) on non-abi3 builds.
float
f64
PyFloat::value