gh-154043: Fix a data race when iterating a shared types.GenericAlias iterator in FT build

[ByteFlowing1337]

• Issue: segfault in ga_iternext: sharing a types.GenericAlias iterator across threads double-frees under free-threading #154043

[ByteFlowing1337]

@serhiy-storchaka Do you mind reviewing this PR? Thanks.

[ByteFlowing1337]

I noticed that there was already a test file named test_types.py written for testing GenericAlias under the FT build, so I moved my test into that file.

But I think its filename is incorrect, it should be test_genericalias.py, corresponding to test_genericalias.py which runs in the normal build.

CC @sobolevn @JelleZijlstra

[serhiy-storchaka]

There is one behavior change here -- when Py_GenericAlias() fails, obj is now consumed anyway. But I think it is fine.

[serhiy-storchaka]

__next__ is supposed to be called immediately after __iter__, in the same thread, of course.

def f(x: *tuple[int]):

So there should not be any race condition on practice. I do not mind using atomic exchange here, but I think that such large and expensive test is not worth it. Remove it.

[serhiy-storchaka]

LGTM. 👍