[clang] [Clang] Implement C++26 P2748R5 "Disallow Binding a Returned Glvalue to a Temporary" (PR #89942)

[A. Jiang via cfe-commits]

================
@@ -8340,8 +8340,17 @@ void Sema::checkInitializerLifetime(const InitializedEntity &Entity,
             << Entity.getType()->isReferenceType() << CLE->getInitializer() << 2
             << DiagRange;
       } else {
-        Diag(DiagLoc, diag::warn_ret_local_temp_addr_ref)
-         << Entity.getType()->isReferenceType() << DiagRange;
+        // P2748R5: Disallow Binding a Returned Glvalue to a Temporary.
+        // [stmt.return]/p6: In a function whose return type is a reference,
+        // other than an invented function for std::is_convertible ([meta.rel]),
+        // a return statement that binds the returned reference to a temporary
+        // expression ([class.temporary]) is ill-formed.
----------------
frederick-vs-ja wrote:

> should we makes `std::is_nothrow_convertible_v<int, Weird>` evaluated as `true`?

I don't see any proper motivation for doing so. Implementations consistently match the design in [P0758R1](https://wg21.link/p0758r1). It's the standard wording that failed to reflect the intent, which is perhaps a defect. (Additionally, the pasted sample implementation would exactly reflect the bug in the `noexcept` operator, if any.)

And IMO `std::is_nothrow_convertible(_v)` should be implementable without intrinsic (and the implementation strategy would be [quite simple](https://en.cppreference.com/w/cpp/types/is_convertible#Possible_implementation)). The `__is_nothrow_convertible` instrinsic should only be meaningful for acceleration of compilation.

https://github.com/llvm/llvm-project/pull/89942