Add a CI check that fails if we are missing a test for a user error

[cburgdorf]

What is wrong?

Some of the errors that the compiler is supposed to present to the user aren't covered by tests.

How can it be fixed

Add a CI check that:

1. extracts a unique part of each user facing error message that the compiler produces
2. Check that the unique part is somewhere present in our snapshots

That should ensure we do not end up in a situation where we have uncovered compiler errors.

[cburgdorf]

Here's an example of code that isn't exercised at all that would have been caught if we had such a check

#762 (comment)

[cburgdorf]

I'm looking back at this because I think it would be really helpful to improve coverage of our tests. There are probably a bunch of user presented errors that we currently do not cover as part of our compile errors test.

I think the simplest thing we could do is to give each error an id and then simply check if all ids are covered in our compile errors tests.

Rust also has ids for the errors and begins each error message with error[<error-id>]. E.g.

error[E0433]: failed to resolve: use of undeclared type `TType`
   --> crates/analyzer/src/traversal/expressions.rs:642:16
    |
642 |             Ok(TType::Mut(field_type(context, inner, field_name, field_span)?).id(context.db()))
    |                ^^^^^
    |                |
    |                use of undeclared type `TType`
    |                help: an enum with a similar name exists: `Type`

For more information about this error, try `rustc --explain E0433`.

We could turn that in a CI check so that it is always guaranteed that each user facing error has a corresponding test.