float=64
precision=double
This avoids confusion with the old `bits=64` option and building for 64-bit CPUs in general.