Support hexfloat in cstdfloat_iostream.hpp

[NAThompson]

No description provided.

[NAThompson]

Main file which exhibits the fix:

#include <iomanip>
#include <iostream>
#include <sstream>

int main()
{
    using boost::float128_t;

    const float128_t x =
        BOOST_FLOAT128_C(3.141592653589793238462643383279502884);

    // Exercise hexfloat output.
    std::stringstream ss;
    ss << std::hexfloat << x;

    std::cout << "hexfloat: " << ss.str() << '\n';

    // Exercise hexfloat input / round-trip.
    float128_t y = 0;
    ss >> y;

    std::cout << "round-trip equal: " << std::boolalpha << (x == y) << '\n';

    // Precision should not affect std::hexfloat.
    std::stringstream a;
    std::stringstream b;

    a << std::hexfloat << std::setprecision(3)  << x;
    b << std::hexfloat << std::setprecision(30) << x;

    std::cout << "precision 3:  " << a.str() << '\n';
    std::cout << "precision 30: " << b.str() << '\n';
    std::cout << "same: " << (a.str() == b.str()) << '\n';

    // Uppercase hexfloat.
    std::stringstream upper;
    upper << std::uppercase << std::hexfloat << x;

    std::cout << "uppercase: " << upper.str() << '\n';

    // Directly parse a hexadecimal floating-point value.
    std::stringstream in("0x1.921fb54442d18469898cc51701b8p+1");

    float128_t z = 0;
    in >> z;

    std::cout << "parsed pi equal: " << (z == x) << '\n';
}

On develop:

➜  math git:(develop) ✗ ./a.out
hexfloat: 3.141593e+00
round-trip equal: false
precision 3:  3.142e+00
precision 30: 3.141592653589793238462643383280e+00
same: false
uppercase: 3.141593e+00
parsed pi equal: false

And on this branch:

➜  math git:(develop) ✗ ./a.out
hexfloat: 3.141593e+00
round-trip equal: false
precision 3:  3.142e+00
precision 30: 3.141592653589793238462643383280e+00
same: false
uppercase: 3.141593e+00
parsed pi equal: false

Closes #1118

[NAThompson]

@ckormanyos: If you could take a quick look . . .