For big numbers (especially integers close to MRB_INT_MAX), float
conversion may be lose precisions. So we try not to convert numbers, but
just compare.