[crater] Normalize opaques to infer vars eagerly in AssocTypeNormalizer

[compiler-errors]

:forgor:

r? @ghost

[compiler-errors]

@bors try

[rustbot]

Some changes occurred to the core trait solver

cc @rust-lang/initiative-trait-system-refactor

[bors]

⌛ Trying commit c353df0 with merge 9b9e20b...

[matthiaskrgr]

owo

[bors]

☀️ Try build successful - checks-actions
Build commit: 9b9e20b (9b9e20bc79549d80ab413f49d990713851172cc1)

[compiler-errors]

@craterbot check

[craterbot]

👌 Experiment pr-120798 created and queued.
🤖 Automatically detected try build 9b9e20b
🔍 You can check out the queue and this experiment's details.

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[craterbot]

🚧 Experiment pr-120798 is now running

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[craterbot]

🎉 Experiment pr-120798 is completed!
📊 853 regressed and 1 fixed (415336 total)
📰 Open the full report.

⚠️ If you notice any spurious failure please add them to the blacklist!
ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[lcnr]

nearly all of the unknown tests are a "concrete type differs from previous defining opaque type use" where one of the hidden types is {type error}. Separate from this PR, we should not emit these errors if the opaques contain a type error.

I think that most of these errors are caused by structurally instantiating opaque types, e.g. coral-0.9.2

pub struct Message {
    pub message: String,
    pub children: Option<Vec<Message>>,
}

impl Message {
    pub fn unroll(&self) -> impl Iterator<Item = &Message> {
        let mut messages = Vec::new();
        messages.push(self);
        if let Some(ref children) = self.children {
            for child in children {
                messages.extend(child.unroll());
                // ^ the returned iter is `opaque::<'fresh>`
            }
        }
        messages.into_iter()
    }
}

unless we want to look through 800 regressions to figure out how many of them are actually the "needs eager inference progress case of recur_rpit().function()" we'd need to also change the old solver to use semantic equality for opaques

[lcnr]

There are definitely some actual regressions here, e.g.

https://github.com/kurnevsky/esxtool/blob/88538ca7c8e068b5e38d4b386eb1bf3d5cede3d0/src/mwscript/parser.rs#L186-L191

and

https://github.com/smol-rs/async-executor/blob/4fbe23af692ac7939d3a281c709c8afd4f0dac95/benches/executor.rs#L76-L80

[compiler-errors]

ok bye bye pr