coherence incorrectly considers `unnormalizable_projection: Trait` to not hold even if it could

[lcnr]

// crate dep
trait IsUnit {}
impl IsUnit for () {}


pub trait WithAssoc<'a> {
    type Assoc;
}

// The two impls of `Trait` overlap
pub trait Trait {}
impl<T> Trait for T
where
    T: 'static,
    for<'a> T: WithAssoc<'a>,
    for<'a> <T as WithAssoc<'a>>::Assoc: IsUnit,
{
}
impl<T> Trait for Box<T> {}
// root crate
use dep::*;

struct Local;
impl WithAssoc<'_> for Box<Local> {
    type Assoc = ();
}

fn impls_trait<T: Trait>() {}

fn main() {
    impls_trait::<Box<Local>>();
}

During coherence for<'a> <Box<T> as WithAssoc<'a>>::Assoc: IsUnit is considered to not hold.
This allows overlapping impls in coherence which is generally considered to be unsound™.

Coherence considers the impls to be disjoint because for<'a> <Box<T> as WithAssoc<'a>>::Assoc: IsUnit, does not hold. When trying to normalize <Box<T> as WithAssoc<'a>>::Assoc ends up ambiguous, we keep the projection around. Coherence then considers for<'a> <Box<T> as WithAssoc<'a>>::Assoc: IsUnit to not have any impl as IsUnit is a local trait.

cc rust-lang/trait-system-refactor-initiative#52

[lcnr]

edit: woops, this comment should have been posted to #105787 instead

[lcnr]

figure out why this detects overlap in the new solver, it is strongly related to #99554