Ensure thread-safety

[LecrisUT]

Are there other static variables that need to be changed?

TODO:

• Test if it does what it says. If it can be done locally, that would be ok as well.
  Maybe a way to do this is to spawn a few thousands workers writing random values to spglib_error_code and re-reading after a delay to confirm it's the same value.
• Check this compiles on intel compiler

[LecrisUT]

@atztogo That was a quick simple fix. Added a simple tester for it as well. (try to remove _Thread_local and it should fail as expected)

[atztogo]

Thanks @LecrisUT. This is interesting. I didn't know about _Thread_local al all. Can it be ignored when using older standard than C11? For many years, spglib had gotten stuck roughly C89 for supporting old microsoft c compiler. Now I don't think we need to support C89, but it made me conservative, and I became to tend to be afraid of new standard. I think what at least we should have to support are the compilers used in conda-forge spglib builds. So I believe there is no issue. Do you have any comment?

[LecrisUT]

So there is also __thread for older gcc compilers. But I don't think it is worth keeping the old standard especially since you can't compile those compilers on currently uaed distros.

I would say a good support track is to stick to what we can check through CI, until there is an issue created for older standard support. So, a common support cycle for this is: conda-forge + containered os (currently Fedora), which usually supports up to 2 previous major releases of the compilers.

If we want an older compiler support, we can make a specific CI job for that based on Ubuntu. But we should state an EOL support for it if we do so.