Completions in macros should be aware of completing multiple idents

[Veykril]

When typing inside a macro we should consider all downmapped positions of the ident we are currently typing out instead of just the first one, that way we get more completions that might make sense.

Example given the following macro:

macro_rules! tuple_to_record_closure {
    ($ident:ident { $( $field:ident ),+ $(,)? }) => {
        |($( $field ),+)| $ident { $( $field ),+ }
    };
}
struct Foo { foo: () }
fn main() {
    tuple_to_record_closure! {
        Foo {
            f$0
        }
    }
}

Typing at the given position should offer the completion of the foo field as the ident is downmapped to a field position as its second usage, currently we only consider the parameter position though.

This also has the potential to create a lot of duplicate completions which have to be filtered out, as well as making completions in macros potentially very slow if we were to stupidly repeat completions for each downmapped token, so this probably requires some architecture changes if we want to pursue this.

cc #11059

[flodiebold]

But shouldn't the completion make sense for all positions? So we'd have to take the intersection? 🤔

[bjorn3]

The completion makes sense in the argument position too, but it isn't suggested as any ident (including a not yet used ident) makes sense there without any preference.

[Veykril]

As bjorn3 said, I think this issue here should be about the union of such completions, while #11059 would be filtering invalid ones out by the fragment capture type. Filtering by expansion makes little sense as the macro input isn't bound to its expansion usage in any way.