libstdc++: Fix std::numeric_limits::lowest() test for strict modes

This test uses std::is_integral to decide whether we are testing an
integral or floating-point type. But that fails for __int128 because
is_integral<__int128> is false in strict modes. By using
numeric_limits::is_integer instead we get the right answer for all types
that have a numeric_limits specialization.

We can also simplify the test by removing the unnecessary tag
dispatching.

libstdc++-v3/ChangeLog:

	* testsuite/18_support/numeric_limits/lowest.cc: Use
	numeric_limits<T>::is_integer instead of is_integral<T>::value.

libstdc++-v3/testsuite/18_support/numeric_limits/lowest.cc

@@ -23,30 +23,20 @@
 // 18.2.1.1 template class numeric_limits
 
 #include <limits>
-#include <type_traits>
 #include <testsuite_hooks.h>
 
 template<typename T>
   void
-  do_test(std::true_type)
+  do_test()
   {
     T limits_min = std::numeric_limits<T>::min();
-    VERIFY( std::numeric_limits<T>::lowest() == limits_min );
-  }
-
-template<typename T>
-  void
-  do_test(std::false_type)
-  {
     T limits_max = std::numeric_limits<T>::max();
-    VERIFY( std::numeric_limits<T>::lowest() == -limits_max );
+    if (std::numeric_limits<T>::is_integer)
+      VERIFY( std::numeric_limits<T>::lowest() == limits_min );
+    else
+      VERIFY( std::numeric_limits<T>::lowest() == -limits_max );
   }
 
-template<typename Tp>
-  void
-  do_test()
-  { do_test<Tp>(typename std::is_integral<Tp>::type()); }
-
 void test01()
 {
   do_test<char>();