Use _Float128 in test for PR118184

The test was failing on x86 because longdouble128 only checks sizeof,
rather than a full 128-bit payload.  Using _Float128 is more portable
and still exposes the original bug.

gcc/testsuite/
	PR target/118184
	* gcc.dg/torture/pr118184.c: Use _Float128 instead of long double.

gcc/testsuite/gcc.dg/torture/pr118184.c

@@ -1,8 +1,8 @@
-/* { dg-do run { target { longdouble128 && lp64 } } } */
+/* { dg-do run { target { float128 && lp64 } } } */
 
 union u1
 {
-  long double ld;
+  _Float128 ld;
   unsigned long l[2];
 };
 
@@ -13,7 +13,7 @@ unsigned long m()
 }
 
 [[gnu::noinline]]
-long double f(void)
+_Float128 f(void)
 {
   union u1 u;
   u.ld = __builtin_nanf128("");