Produce a rational number for any real numeric argument x. This is more efficient than rationalize, but it assumes that floating-point is completely accurate, giving a result that is not as pretty.