qfloat16: make native std::numeric_limits<qfloat16> constexpr

In C++23 with std::float16_t present (QFLOAT16_IS_NATIVE), the
_limit_xxxx() methods in qfloat16 need to set the native type, not rely
on the union trick because that isn't allowed in constexpr mode.

                    in ‘constexpr’ expansion of ‘operator<=>(Max, std::numeric_limits<double>::max())’
qfloat16.h:209:5:   in ‘constexpr’ expansion of ‘compareThreeWay((* & lhs), ((double)rhs))’
qfloat16.h:209:5:   in ‘constexpr’ expansion of ‘(& lhs)->qfloat16::operator NativeType()’
    error: accessing ‘qfloat16::<unnamed union>::nf’ member instead of initialized ‘qfloat16::<unnamed union>::b16’ member in constant expression

Pick-to: 6.7
Change-Id: I01ec3c774d9943adb903fffd17b8b6ceed6ef9e2
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>

src/corelib/global/qfloat16.h

@@ -118,7 +118,13 @@ private:
         NativeType nf;
 #endif
     };
-    constexpr inline explicit qfloat16(Wrap nibble) noexcept : b16(nibble.b16) {}
+    constexpr inline explicit qfloat16(Wrap nibble) noexcept :
+#if QFLOAT16_IS_NATIVE && defined(__cpp_lib_bit_cast)
+        nf(std::bit_cast<NativeType>(nibble.b16))
+#else
+        b16(nibble.b16)
+#endif
+    {}
 
     Q_CORE_EXPORT static const quint32 mantissatable[];
     Q_CORE_EXPORT static const quint32 exponenttable[];