Rollup merge of #91412 - compiler-errors:issue-86035, r=oli-obk

Improve suggestions for importing out-of-scope traits reexported as `_`

1. Fix up the `parent_map` query to prefer visible parents that _don't_ export items with the name `_`.
    * I'm not sure if I modified this query properly. Not sure if we want to check for other idents than `kw::Underscore`.
    * This also has the side-effect of not doing BFS on any modules re-exported as `_`, but I think that's desirable here too (or else we get suggestions for paths like `a::_::b` like in [this doctest example](https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=d9505ea45bb80adf40bb991298f952be)).
2. Bail in `try_print_visible_def_path` if the `def_id` is re-exported as `_`, which will fall back to printing the def-id's real (definition) path.
    * Side-effect of this is that we print paths that are not actually public, but it seems we already sometimes suggest `use`ing paths that are private anyways. See [this doctest example](https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=bad513ed3241f8ff87579eed8046ad10) for demonstration of current behavior.
3. Suggest a glob-import (for example `my_library::prelude::*`) if the trait in question is only pub-exported as `_`, as a fallback.
    ```
    use foo::bar::prelude::*; // trait MyTrait
    ```
    * I think this is good fallback behavior to suggest instead of doing nothing. Thanks to the original issue filer for suggesting this.

I was somewhat opinionated about behaviors in this PR, and I'm totally open to limiting the impact of my changes or only landing parts of this. Happy to receive feedback if there are better ways to the same end.

Fixes #86035

0 files changed, 0 insertions, 0 deletions