gh-142254: add @cpython_only to tests relying on deepcopy implementation

[Bobronium]

• Issue: Test deepcopy() memo behavior without inspecting internals #142254

Replaces gh-142255

[picnixz]

@serhiy-storchaka What do you think of this one? strictly speaking, we're reducing test coverage of other implementations but I don't know if @cpython_only applies to stdlib in pure Python. Strictly speaking, our pickle implementation deviates from the docs themselves so it's strictly speaking, a cpython-only thing, but I'm not sure we're really consistent actually. I'm also fine with not changing anything.

[serhiy-storchaka]

I do not think this is an appropriate use of cpython_only. The implementation is not CPython specific.

[picnixz]

So for you cpython_only only applies to the interpreter's implemenation and not the modules in Lib right? I'm actually fine with marking it as cpython_only but I don't like the fact that other implementations that use the same code would then have a smaller test coverage. I think it's then better to not mark it as an implementation detail even if we disagree with the docs a bit, but and others will need to write their own tests instead (but I don't think we broke any major implementation right?) WDYT?

[serhiy-storchaka]

If there is a major implementation which fails tests for good reason, they can be marked as CPython-only. It can be a code in Lib if it depends on CPython-specific behavior or modules.

But this is not the case.

[picnixz]

Ok, let's close this one. Sorry @Bobronium but there isn't enough reason to change our current code for that.