normalizing opaques while proving implied bounds

[lcnr]

This is unsound with the new solver with rust-lang/rust#136074.

I am using a OutlivesComponent::Alias as these use equality on the computed region constraints graph. When computing the implied bounds, the hidden type of impl Sized still has unconstrained regions, so region outlive bounds don't get added to the free_region_relations.

use std::any::Any;

struct Outlives<'a, T>(Option<&'a T>);
trait Trait {
    type Assoc;
}

impl<T> Trait for T {
    type Assoc = T;
}

// Computing the implied bounds for `foo` normalizes `impl Sized` to
// `Outlives::<'static, <T as Trait>::Assoc>`, adding the implied bound
// `<T as Trait>::Assoc: 'static`.
//
// The caller does not have to prove that bound.
fn foo<T: Trait>(x: <T as Trait>::Assoc) -> (Box<dyn Any>, impl Sized) {
    (Box::new(x), Outlives::<'static, <T as Trait>::Assoc>(None))
}

fn main() {
    let string = String::from("temporary");
    let (any, _proof) = foo::<&str>(string.as_str());
    drop(_proof);
    drop(string);
    println!("{}", any.downcast_ref::<&str>().unwrap());
}

[compiler-errors]

Not sure how we're gonna do this, especially b/c now with TypingEnv being associated with an inference context it's not so easy to just swap out the reveal mode to something that keeps the opaques opaque.

[compiler-errors]

Actually this is doable