Fix recursion issues

[jobh]

Trying to finally kill that recursion zombie.

from_type didn't properly detect recursion on types that went through types.from_typing_type. Fixed by moving the recurse-guard list to a ContextVar. (yep, I should have taken your advice on that in the first place @Zac-HD)

Additionally:

• when nicerepr encounters an error (RecursionError for example), convert it to a warning and return a string anyway. This is what made it easy to identify the problem in the first place,
• hard-code the stack depth increment as 2000 to avoid inconsistency between test runners (pytest seems to bump it from 1000 to 3000 during test execution, for example, and other packages may interfere),
• some other minor stuff which I'll comment in the code.

Closes: #3525

[jobh]

The example in #3026 (comment), actually didn't work now, because Optional[RecursiveClass].__module__ is typing. (RecursiveClass|None).__module__ is types, so that happened to work. And builtins is probably also unnecessary because relevant builtins are registered in _global_type_lookup anyway.

Long story short: Unless there's a reason I don't see, just remove the __module__ check and everything is fine.

[jobh]

Just FYI, this was the cause of the ghostwriter failures (now fixed).

[jobh]

@Zac-HD I rebased this PR without the fluff, but including some cherry-picks from #3688.

[Zac-HD]

I'm looking forward to getting this in - see comments below, add release notes, and it looks like we'll need to do something about os.environ in the ghostwriter too 🤔

[jobh]

I'm looking forward to getting this in - see comments below, add release notes, and it looks like we'll need to do something about os.environ in the ghostwriter too thinking

I think I addressed everything (except ghostwriter), let's see how the tests turn out. Will revisit if not, but it may be a few days again

[jobh]

Synced and rebased to fix conflict

[Zac-HD]

Let me know if you want me to dig in and offer some suggestions for or to push a fix for the ghostwriter issue; by default I'm happy to leave it to your judgement.

[jobh]

Let me know if you want me to dig in and offer some suggestions for or to push a fix for the ghostwriter issue; by default I'm happy to leave it to your judgement.

I committed a change to ghostwriter to just undo the earlier (new) deferring. I'm not familiar enough with it to tell whether this is sufficiently precise or robust

691b021

[jobh]

Hm, this will actually share the same list across threads (contexts), won't it. I'll push a fix but it will add a bit of ugly

[Zac-HD]

Ah, the merge conflict is from this ongoing merge and should resolve itself momentarily.

Content looks great though, so I'll merge as soon as that previous release finishes!