[ty] Make tuple subclass constructors sound

[AlexWaygood]

Summary

#18987 only partly implemented the special-casing required to make tuple instantiations sound. While that PR made instantiations of tuple itself sound, it did not make instantiations of tuple subclasses sound, as the way in which that PR added the special casing did not mean that the synthesized constructor signatures would be automatically inherited by subclasses. This PR fixes that by moving the special casing to a lower level, inside ClassType::own_class_member.

Test Plan

Mdtests added. Some existing mdtests also had to be slightly modified because they were unsoundly constructing instances of tuple subclasses.

[github-actions]

mypy_primer results

No ecosystem changes detected ✅
No memory usage changes detected ✅

[AlexWaygood]

The primer hits seem to mostly be people doing list(map(tuple, iterable_of_iterables)):

• https://github.com/scipy/scipy/blob/8dd9d43e62f3f6401599ced3917835d4458cece4/scipy/sparse/_construct.py#L1286
• https://github.com/scipy/scipy/blob/8dd9d43e62f3f6401599ced3917835d4458cece4/scipy/spatial/tests/test_qhull.py#L25
• https://github.com/scipy/scipy/blob/8dd9d43e62f3f6401599ced3917835d4458cece4/scipy/spatial/tests/test_qhull.py#L27
• https://github.com/more-itertools/more-itertools/blob/89172166167a5323c19ecb45fee49f191ea49bcd/more_itertools/more.py#L4164

[carljm]

The new diagnostics look weird, though, because I would expect <class 'tuple'> to be assignable to (Unknown, /) -> Unknown.

[AlexWaygood]

The new diagnostics look weird, though, because I would expect <class 'tuple'> to be assignable to (Unknown, /) -> Unknown.

yeah, that's why it's still in draft :P I'm looking into it!

[AlexWaygood]

Minimal repro: this still passes on this branch:

from typing import Callable, Any

def f(x: Callable[..., Any]): ...

f(tuple)

but this no longer does:

from typing import Callable, Any

def f(x: Callable[[Any], Any]): ...

f(tuple)

[AlexWaygood]

The reason seems to be that this .into_callable(db) call is now returning Callable[[], tuple[Unknown, ...]] for <class 'tuple'>:

ruff/crates/ty_python_semantic/src/types.rs

Lines 1420 to 1422 in fcdffe4

	(_, Type::Callable(_)) => self
	.into_callable(db)
	.is_some_and(|callable| callable.has_relation_to(db, target, relation)),

[AlexWaygood]

Oh, I think the bug is just astral-sh/ty#760 again, but for __new__ this time rather than __init__

[AlexWaygood]

Great, 3838203 fixed it! This now has a clean primer report.