runner: re-raise pytest.exit inside ExceptionGroups (fixes #13650)

[gomri15]

Fix issue: 13650, fixes regression when adding exception group which meant pytest raises in teardown wouldn't stop execution

Scope note: This PR intentionally does not handle nested ExceptionGroups (EGs containing EGs). That can be discussed in a follow-up if desired.

Handle pytest.exit() in exception group by searching in exception group if pytest.exit() found and reraising it.

Rationale:
Before Python 3.11 (no EGs), CallInfo.from_call re-raised certain exceptions (e.g., Exit, KeyboardInterrupt) so they bypassed reporting. With EGs, those exceptions could be inside a group, and the simple isinstance(excinfo.value, reraise) check no longer triggered. The session would continue to the next test instead of stopping. This PR restores the original contract in an EG world, without introducing new flows or flags.

Fixes #13650

[nicoddemus]

Thanks @gomri15 for the PR!

Left some small suggestions, please take a look.

[gomri15]

@nicoddemus You guys squash all the changes or leave them?, i assumed squash but want to make sure.

[gomri15]

@nicoddemus Is this something were planning to eventually merge you think?.
or is there some process in which more obscure bug fixes are only merged nearer version release?

[bluetech]

The change looks good and clean to me, thanks!

I think we need to make sure we agree on the semantics we want though, I will post a comment on the issue.

You guys squash all the changes or leave them?, i assumed squash but want to make sure.

If the author makes the commits "atomic" i.e. each commit stands on its own (contains one logical change, passes tests etc.), then we merge. That's how I like to do it, but most people don't bother with this, instead the PR should be one single commit, then we squash. You can work whichever way you feel comfortable.

[bluetech]

Scope note: This PR intentionally does not handle nested ExceptionGroups (EGs containing EGs). That can be discussed in a follow-up if desired.

Should be fine in a follow up, but would be good to see if we can make it work, if only for pedantic correctness.

I think we could do it like this:

1. Use eg.group(reraises) (no need for reraise_types)
2. If not None, flatten it and return the first exception

As for flattening, there seems to not be a built-in way to do it, but I see there is a proposed PEP adding leaf_exceptions() and it has an implementation snippet, so we can perhaps just throw it in compat.py and use it. Or maybe just hand-roll a simple version of it that just returns the first exception.