Post lazy normalization for consts cleanup (#71973)

[lcnr]

While #71973 implements lazy normalization for constants,
there are still some bugs we need to deal with afterwards.

This issue is a collection of things not explicitly mentioned with FIXME(lazy_normalization_consts) so we don't forget anything.

[lcnr]

done

Add src/test/ui/const-generics/issues/issue-69654.rs back in, which causes an unrelated stack overflow once lazy normalization is enabled.

#![feature(const_generics)]
#![allow(incomplete_features)]

trait Bar<T> {}
impl<T> Bar<T> for [u8; T] {}
//~^ ERROR expected value, found type parameter `T`

struct Foo<const N: usize> {}
impl<const N: usize> Foo<N>
where
    [u8; N]: Bar<[(); N]>,
{
    fn foo() {}
}

fn main() {
    Foo::foo();
}

Also add the following similar test, which should compile but also causes a stack overflow.

// run-pass
#![feature(const_generics)]
#![allow(incomplete_features, unused_braces)]

trait Bar<T> {}
impl<T> Bar<T> for [u8; {7}] {}

struct Foo<const N: usize> {}
impl<const N: usize> Foo<N>
where
    [u8; N]: Bar<[(); N]>,
{
    fn foo() {}
}

fn main() {
    Foo::foo();
}

[lcnr]

Taken from a comment by @nikomatsakis

// run-pass
#![feature(const_generics)]
#![feature(lazy_normalization_consts)]
#![allow(incomplete_features)]

// try to devise a test for higher-ranked case
//
// e.g., for<'a> EqZero<<() as Zero<'a>::C>>
//
//

trait EqZero<const C: usize> { }

impl EqZero<0> for () { }

trait Zero<'a> {
    const C: usize;
}

impl<'a> Zero<'a> for () {
    const C: usize = 0;
}

fn test_me<T>()
where for<'a> T: EqZero<{<() as Zero<'a>>::C}>
{
}

fn main() {
    test_me::<()>();
}

However, that test continues to ICE even with these changes, due to some assertion in type checking. So maybe it's fine, and we should just move that test to a follow-up issue. Anyway, this code as it is written here is the code we want eventually -- we just have to reconcile some details of the leak-check first.

[varkor]

Linking the relevant Zulip conversation here, which is helpful.