we never try to normalize unused alias definitions

[lcnr]

causing check builds to pass

affected tests

• we never try to normalize unused inherent associated types
  • tests/ui/associated-inherent-types/normalization-overflow.rs
• we never try to normalize lazy type aliases
  • tests/ui/infinite/infinite-type-alias-mutual-recursion.rs

struct T;

impl T {
    type This = Self::This; //~ ERROR overflow evaluating associated type `T::This`
}

fn main() {}

[lcnr]

we also do not eagerly try to normalize struct fields

trait Trait {
    type Diverges<D: Trait>;
}

impl<T> Trait for T {
    type Diverges<D: Trait> = D::Diverges<D>;
}

struct Foo {
    field: Box<<u8 as Trait>::Diverges<u8>>,
}

[BoxyUwU]

The same is also true of defaulted type parameters:

trait Trait {
    type Diverges<D: Trait>;
}

impl<T> Trait for T {
    type Diverges<D: Trait> = D::Diverges<D>;
}

struct Bar<T: ?Sized = <u8 as Trait>::Diverges<u8>>(Box<T>);

Though this is not a new solver regression as the old solver also fails to emit an error here