It avoids a lot of repetition.

…rors

resolve: Unload speculatively resolved crates before freezing cstore

Name resolution sometimes loads additional crates to improve diagnostics (e.g. suggest imports).
Not all of these diagnostics result in errors, sometimes they are just warnings, like in rust-lang#117772.

If additional crates loaded speculatively stay and gets listed by things like `query crates` then they may produce further errors like duplicated lang items, because lang items from speculatively loaded crates are as good as from non-speculatively loaded crates.
They can probably do things like adding unintended impls from speculatively loaded crates to method resolution as well.
The extra crates will also get into the crate's metadata as legitimate dependencies.

In this PR I remove the speculative crates from cstore when name resolution is finished and cstore is frozen.
This is better than e.g. filtering away speculative crates in `query crates` because things like `DefId`s referring to these crates and leaking to later compilation stages can produce ICEs much easier, allowing to detect them.

The unloading could potentially be skipped if any errors were reported (to allow using `DefId`s from speculatively loaded crates for recovery), but I didn't do it in this PR because I haven't seen such cases of recovery. We can reconsider later if any relevant ICEs are reported.

Unblocks rust-lang#117772.

…oli-obk

Make it so that async-fn-in-trait is compatible with a concrete future in implementation

There's no technical reason why an AFIT like `async fn foo()` cannot be satisfied with an implementation signature like `fn foo() -> Pin<Box<dyn Future<Output = ()> + 'static>>`.

We rejected this previously because we were uncertain about how AFITs worked with refinement, but I don't believe this needs to be a restriction any longer.

r? oli-obk

…rrors

hir: Make sure all `HirId`s have corresponding HIR `Node`s

And then remove `tcx.opt_hir_node(hir_id)` in favor of `tcx.hir_node(hir_id)`.

match lowering: consistently lower bindings deepest-first

Currently when lowering match expressions to MIR, we do a funny little dance with the order of bindings. I attempt to explain it in the third commit: we handle refutable (i.e. needing a test) patterns differently than irrefutable ones. This leads to inconsistencies, as reported in rust-lang#120210. The reason we need a dance at all is for situations like:

```rust
fn foo1(x: NonCopyStruct) {
    let y @ NonCopyStruct { copy_field: z } = x;
    // the above should turn into
    let z = x.copy_field;
    let y = x;
}
```

Here the `y ```````@```````` binding will move out of `x`, so we need to copy the field first.

I believe that the inconsistency came about when we fixed rust-lang#69971, and didn't notice that the fix didn't extend to refutable patterns. My guess then is that ordering bindings by "deepest-first, otherwise source order" is a sound choice. This PR implements that (at least I hope, match lowering is hard to follow 🥲).

Fixes rust-lang#120210

r? ```````@oli-obk``````` since you merged the original fix to rust-lang#69971
cc ```````@matthewjasper```````

GVN: also turn moves into copies with projections

Fixes rust-lang#120613

docs: also check the inline stmt during redundant link check

Fixes rust-lang#120444

This issue was brought about by querying `root::webdavfs::A`, a key that doesn't exist in `doc_link_resolutions`. To avoid a panic, I've altered the gating mechanism to allow this lint pass to be skipped.

I'm not certain if this is the best solution. An alternative approach might be to leverage other info from the name resolutions instead of `doc_link_resolutions`. After all, all we need is to get the resolution from a combination of `(module, name)`. However, I believe they would yield the same outcome, both skipping this lint.

…ompiler-errors

exhaustiveness: Prefer "`0..MAX` not covered" to "`_` not covered"

There was an exception when reporting integer ranges as missing, it's been there for as long as I can remember. This PR removes it. I think it's nicer to report "`0..MAX` not covered" than "`_` not covered". This also makes it consistent with enums, where we report individual enum variants in this case (as showcased in the rest of the `empty-match.rs` test).

r? ``@estebank``

…, r=compiler-errors

Add `SubdiagnosticMessageOp` as a trait alias.

It avoids a lot of repetition.

r? matthewjasper

…r-errors

improve pretty printing for associated items in trait objects

* Don't print a binder in front of associated items, because it's not valid syntax.
  * e.g. print `dyn for<'a> Trait<'a, Assoc = &'a u8>` instead of `dyn for<'a> Trait<'a, for<'a> Assoc = &'a u8>`.
* Don't print associated items that are implied by a supertrait bound.
  * e.g. if we have `trait Sub: Super<Assoc = u8> {}`, then just print `dyn Sub` instead of `dyn Sub<Assoc = u8>`.

I've added the test in the first commit, so you can see the diff of the compiler output in the second commit.