sizeComplete pragma to workaround C++ Atomic [backport]

[metagn]

fixes #24159, refs #23761, refs #15928, alternative to #24166

As an alternative to #24166, instead of disabling typeinfo generation on all completeStruct imported C++ types, add a new pragma sizeComplete that acts the same as completeStruct except generating typeinfo, and use this pragma for Atomic[T]. This seems like the most clearly temporary solution, I also considered a noTypeInfo pragma but it seemed like it would overstay its welcome.

[Araq]

No. Let's fix Atomic[T] instead. It shouldn't need the raw field.

[arnetheduck]

the easy fix here is to c atomics also on c++

[metagn]

the easy fix here is to use c atomics also on c++

Lol. Maybe we can add a define for this regardless but it won't work out of the box since C++ doesn't have stdatomic.h etc

[arnetheduck]

C++ doesn't have

tbh, if I were to pick a C implementation, I'd use stdatomics.h as a fallback and first and foremost use any compiler-specific C implementation with intrinsics as these have been around the longest and are therefore the most compatible with older versions of such compilers.

only when the compiler is unknown (ie not gcc/clang/msvc more or less) would I "fall back" on stdatomics (which is part of c11 but indeed only becomes part of c++ in c++23.

[metagn]

With #24207 + #24209 the C++ atomic methods are now opt-in, the C11 atomic functions which are now used are in the C++11 standard for the atomic header. It's definitely nicer in the sense that the difference between the backends is limited.