[red-knot] No cyclic-class-def diagnostics for subclasses of cyclic classes

[wooly18]

Summary

Resolves #14223

Test Plan

cargo nextest run

[github-actions]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

[carljm]

This looks great, thank you for the contribution!

The algorithm looks good, my only comments are on naming and clarity. Apologies, this is slightly bike-sheddy. I think the Option<bool> return type (and its semantics: Some if there is a cycle, else None, inner bool is true if the class itself is a participant in the cycle) are too non-obvious. I would prefer an explicit InheritanceCycle enum with variants Inherited and Participant (with doc comments clarifying the meaning of each), and return Option<InheritanceCycle> instead. We can define a method like is_participant() to make the usage of this enum in the if test for the diagnostic a bit more ergonomic.

I'm also not sure that is_involved_in_cyclic_definition adds enough clarity to justify its length. I'm guessing you renamed from is_cyclically_defined because that seems to imply the definition of this class itself is part of the cycle? I agree with that. I'd suggest maybe just inheritance_cycle() as the method name, leading to usages looking like class.inheritance_cycle().is_some(), which reads pretty naturally?

@AlexWaygood wrote this code and may also have thoughts, not sure if he'll be able to offer them before Monday, though.

[wooly18]

Thanks for the feedback! The name changes sound good to me and I'm happy to make them soon.