should survey uses of `#[rustc_error]` and see if they can be replaced with `// compile-pass`

[pnkfelix]

Discussing some of the difficulties with #54528 with @nikomatsakis led them to remind me that we originally introduced #[rustc_error] because we didn't have a way to express // compile-pass to the test infrastructure.

But now we do

and it would make sense for us to try to migrate instances of #[rustc_error] to use // compile-pass when possible.

(I think the main case where it is not possible are instances of // revisions where at least one of the revisions compiles successfully and another does not.)

[pnkfelix]

i guess they should also use // skip-trans or whatever it is that skips running LLVM

update: // skip-codegen; cc #49321