Add performance test comparing double to mp_number<float_backend<double>>. Fix typo in docs and regenerate.