Stl: <complex> asinh benefit from constexpr?

[SunnyWar]

Currently, the code is written:
````
static _Ty asinh(_Ty _Left) {
constexpr _Ty _Ln2 = 0.69314718055994530941723212145817658L;

    bool _Neg = _Left < 0;
    _Ty _Ans;

    if (_Neg) {
        _Left = -_Left;
    }

    if (_Left < 2 / _Flt_eps()) {
        _Ans = log1p(_Left + _Left * _Left / (1 + sqrt(_Left * _Left + 1)));
    } else {
        _Ans = log(_Left) + _Ln2;
    }

    return static_cast<_Ty>(_Neg ? -_Ans : _Ans);
}

````
I tried the following in compiler explorer and it appeared to generate slightly better assembly.

````
static _Ty asinh(_Ty _Left) {
constexpr _Ty _Ln2 = 0.69314718055994530941723212145817658L;
constexpr _Ty boundary = 2 / std::numeric_limits<_Ty>::epsilon();

    bool _Neg = _Left < 0;
    double _Ans;

    if (_Neg) {
        _Left = -_Left;
    }

    if (_Left < boundary) {
        _Ans = log1p(_Left + _Left * _Left / (1 + sqrt(_Left * _Left + 1)));
    } else {
        _Ans = log(_Left) + _Ln2;
    }

    return static_cast<double>(_Neg ? -_Ans : _Ans);
}

````

[StephanTLavavej]

Since you’re proposing a specific code change, please submit this as a pull request so you can sign our CLA. Thanks!

[StephanTLavavej]

Closing this for now, but we're still open to reviewing a PR.