Implement is_floating_point and is_arithmetic for float128

[Charles Karney <charles@…>]

The following program

#include <iostream>
#include <random>
#include <boost/multiprecision/float128.hpp>

int main() {
  unsigned s = std::random_device()();
  std::mt19937 g(s);
  std::uniform_real_distribution<boost::multiprecision::float128> dis;
  std::cout << dis(g) << "\n";
}

fails to compile with g++ 6.3.1 due to this static assertion failure

/usr/include/c++/6.3.1/bits/random.h:160:2: error: static assertion failed: template argument not a floating point type
  static_assert(std::is_floating_point<_DInputType>::value,
  ^~~~~~~~~~~~~

[John Maddock]

Sorry, no can do: type_traits work on the what kind of type an object is, not how they behave. float128 is a class type (and responds true to is_class), it genuinely should not respond true to is_arithmetic etc.

You code above should work fine if you use boost::mt19937 etc rather than the std versions.

It's probably worth while filing a gcc/libstdc++ bug report as well since the assert they are using could be replaced by one that uses numeric_limits<> to verify a type behaves as a floating point type - and numeric_limits is one trait we are allowed to specialize (and do so).