Allow constraining opaque types during subtyping in the trait system

[oli-obk]

Previous attempt: #123979

Sometimes we don't immediately perform subtyping, but instead register a subtyping obligation and solve that obligation when its inference variables become resolved. Unlike immediate subtyping, we currently do not allow registering hidden types for opaque types. This PR also allows that.

[lcnr]

A minimized test:

fn foo() -> impl Default {
    let x = Default::default();
    // add `Subtype(?x, ?y)` obligation
    let y = x;

    // Make a tuple `(?x, ?y)` and equate it with `(impl Default, u32)`.
    // For us to try and prove a `Subtype(impl Default, u32)` obligation,
    // we have to instantiate both `?x` and `?y` without any
    // `select_where_possible` calls inbetween.
    let tup = &mut (x, y);
    let assign_tup = &mut (foo(), 1u32);
    tup = assign_tup;
    1u32
}

fn main() {}

I think this change is clearly desirable, even if it did no go through an FCP. From what I can tell it also doesn't block any other work, as I want us to convert all places to use DefineOpaqueTypes::Yes before stabilizing ATPIT. I would like us to FCP this.

While I think it would have been fine to FCP merge this without first reverting this PR, we would only have 1 week to go through the FCP to make it in time for its stabilization. Given that we're missing some existing uncertainty wrt the impact of this change, please revert #123979 and remerge it after a types FCP.

[compiler-errors]

@bors r+

[bors]

📌 Commit 1e9cd07 has been approved by compiler-errors

It is now in the queue for this repository.

[jieyouxu]

UI tests might need rebless #126653 (comment)
@bors r-

[oli-obk]

@bors r=compiler-errors

[bors]

📌 Commit ba4510e has been approved by compiler-errors

It is now in the queue for this repository.

[bors]

⌛ Testing commit ba4510e with merge 3186d17...

[fmease]

It's already merged, bors. What the hell?
@bors r-

Edit: We currently have three full-build pending PRs, omg.

[fmease]

@bors retry r-