fix: set local_module before elaborating each trait

[asterite]

Description

Problem

Resolves #6479

Summary

This was actually more than just a visibility bug: it could happen that we just elaborated a trait in a nested module, and local_module ended up in that nested module; then finding types for where clauses would find them even if they were normally unreachable. For example this code compiled fine:

mod foo {
    pub trait Foo {}
}

pub trait Bar<T>
where
    T: Foo, // Here Foo was incorrectly found
{}

Additional Context

I wonder if directly setting local_module should be forbidden, and instead we should always use replace_module.

Documentation

Check one:

• No documentation needed.
• Documentation included in this PR.
• [For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[jfecher]

Good catch. I wonder if we're missing this for the other collect cases as well?

Also, should we set the current file or anything else as well? I suppose it is the call to resolve_trait_constraints that is why we require the local module here. Makes me wonder if we have any missed assumptions elsewhere.

[asterite]

I'll merge this now but next week I'll review all locations and see if it can be refactroed into something else.