Several tests with `_testcapi.set_nomemory` miss some skips

[sobolevn]

All of them must have:

• support.cpython_only
• Skip if _testcapi is missing
• Skip if Py_TRACE_REFS is not supported

Refs #152229 (review)

I will send a PR.

Linked PRs

• gh-152236: Fix skips on _testcapi.set_nomemory tests #152253
• [3.15] gh-152236: Fix skips on _testcapi.set_nomemory tests (GH-152253) #152281
• [3.14] gh-152236: Fix skips on _testcapi.set_nomemory tests (GH-152253) #152282
• [3.13] gh-152236: Fix skips on _testcapi.set_nomemory tests (GH-152253) #152286

[StanFromIreland]

support.cpython_only

I think this is wrong, I know at least RustPython is trying to implement _testcapi.

[StanFromIreland]

Oh and how could I (almost) forget! We provide it in PyPy.

[sobolevn]

_testcapi.set_nomemory is very specific to the allocations count. So, it can't be tested on other implementations reliably.

[StanFromIreland]

_testcapi.set_nomemory is very specific to the allocations count. So, it can't be tested on other implementations reliably.

It can't be tested on CPython reliably either, free-threading builds, JIT builds, various platforms all do different things. I don't think we should mark tests as CPython only simply because they use _testcapi.

[sobolevn]

Yes, they can be ~reliably tested in CPython :)
Because they are right now. They execute on different envs, diffent builds.

But, we can't enforce tests of the internal details and memory management tests on other implementations. Other implementations will use different memory allocation patterns.