[red-knot] add narrowing support in match class patterns

[carljm]

The goal of this issue is to make this mdtest pass:

def get_bool() -> bool: ...
def get_object() -> object: ...
class A: ...
class B: ...

x = get_object()

reveal_type(x)  # revealed: object

match x:
    case A():
        reveal_type(x)  # revealed: A
    case B():
        reveal_type(x)  # revealed: B

reveal_type(x)  # revealed: object

This should just require filling in the TODO for MatchClass at

ruff/crates/red_knot_python_semantic/src/types/narrow.rs

Line 232 in 83651de

None // TODO

We don't need to worry about arguments yet; the goal here is just to narrow based on the implicit isinstance check performed by the class pattern.

See https://docs.python.org/3/reference/compound_stmts.html#class-patterns for the documentation of the language feature.

[InSyncWithFoo]

I looked into this and got a major block:

match pattern.pattern(self.db).node() {
    // ...
    ast::Pattern::MatchClass(PatternMatchClass { cls, .. }) => {
        let inference = infer_scope_types(self.db, self.scope());  // This panics.
    }
    // ...
}

The last logged message is Box<dyn Any>.

The problem appears to be reproducible with just:

```py
x = object()

match x:
    case int():
        x  # Or any name
```

This comment seems relevant:

// SAFETY: we should always have a symbol for every Name node.

[carljm]

The inscrutable Box<dyn Any> error message indicates a Salsa query cycle. In this case it would be due to calling infer_scope_types in type narrowing, when in turn type narrowing may well have been called by infer_scope_types on that same scope.

For expressions that function as narrowing predicates, we will usually need to add them as standalone expressions when building the semantic index, so we can query their type independently. Though since match patterns don't use Expr but rather ComparableExpr, this will probably require adding some new infrastructure for standalone ComparableExpr.

(Note this task is assigned to me, it's not up-for-grabs at the moment.)

[sharkdp]

Just a quick note that I touched some of the relevant code in #14759. So maybe contact me before starting with this task (in case #14759 has not been merged in the meantime).