The IEEE standard comes with a `squareRoot`operation with guaranteed-precision results. Would be nice to support that in apfloat as well. :)