TSRM: Replace TSRM_TLS by C11’s thread_local

[TimWolla]

No description provided.

[TimWolla]

Appears to be blocked on #20064.

[nielsdos]

This duplicates some code regarding the includes, so I don't particularly like this. Imo it adds more complexity

[dstogov]

I wouldn't do this. I would prefer to keep TSRM as an abstraction layer.

[TimWolla]

@nielsdos The duplication arises from the TLS testing files that Arnaud added for static OPcache. Other than that the change remains localized to TSRM.

@dstogov TSRM still remains an abstraction layer for most of the thread-safety support. But it seems useful to me to rely on standardized functionality where possible.

---

A good think that this PR pointed out: We were apparently not yet building with C11 on Windows. It makes sense to me to ship that independently. Unfortunately my attempt to change that failed, due to ext/intl which needs C++. I'm not seeing how to easily add the /std: flag for everything when not overridden. And I also don't have any Windows experience (or even a Windows machine available). Perhaps @cmb69 can provide some insight?

[cmb69]

C++. I'm not seeing how to easily add the /std: flag for everything when not overridden.

Generally, it might be sensible to distinguish between C and C++ code for Windows/MSVC (like we do on POSIX platforms). Currently, our build system is totally agnostic to that; everything is passed to MSVC, and the compiler decides whether it is C or C++ (based on the file extension), and there is only one set of flags (CFLAGS).

That said, I don't think this will help in this case. If ext-intl (or any other C++ extension) would be build for ZTS, the build would fail, since we can't have C11 atomics in C++ code to my knowledge.

Why not stick with TSRM_TLS anyway?