support typing.TypeAlias (PEP 613)

[carljm]

If a name is annotated as typing.TypeAlias, the RHS should be evaluated as a type expression.

[sharkdp]

A minor note here: Not that this needs to be a blocker, but if we implement this before fixing the recursive types problem, we'll have to move ~10 projects to "bad.txt" (ref), and we probably have to expect more panics on user projects as well.

[AlexWaygood]

A minor note here: Not that this needs to be a blocker, but if we implement this before fixing the recursive types problem, we'll have to move ~10 projects to "bad.txt" (ref), and we probably have to expect more panics on user projects as well.

I'm sort-of surprised that's "all"! Typeshed uses recursive type aliases for some very fundamental functions in the stdlib: https://github.com/astral-sh/ruff/blob/e2d96df50192803abe63cb0865b447c2552c2eff/crates/ty_vendored/vendor/typeshed/stdlib/builtins.pyi#L1533. I'm somewhat uneasy about implementing support for TypeAlias unless we're confident that we don't (and won't) panic on typeshed's annotations there.

(Even if we don't currently panic on that alias after support for TypeAlias has been added, is there a risk that there's another missing feature that means we don't understand that alias fully right now which, when implemented, suddenly causes us to start panicking?)