RES: prefer trait methods before private inherent methods

[vlad20012]

Improves the #5661 fix: now resolve it to visible trait impl instead of private inherent impl:

use foo::{Foo, Trait};

mod foo {
    pub struct Foo;
    impl Foo {
        // Private
        fn get(&self) { println!("struct"); }
    }

    pub trait Trait {
        fn get(&self);
    }
    impl Trait for Foo {
        fn get(&self) { println!("trait"); }
    }    //X
}

fn main() {
    let f = foo::Foo;
    f.get();
    //^
}

c.c. @Kobzol

[Kobzol]

Even better :) Maybe we could keep the old test with

impl Foo {
    pub fn get(&self) { println("struct"); }
}

?

[vlad20012]

@Kobzol Hmm yep, Let's keep it

[vlad20012]

@Kobzol Oh, actually there isn't a multiresolve because inherent impl wins

[Kobzol]

Oh, ok, even better :)

[vlad20012]

@Undin the PR awaiting the regression testing repairs :)

[Undin]

@vlad20012 as we discussed in person, regression-2 is only about stdlib and yeah, it's broken currently for some reason. All other regression tests work as expected so looks like you don't need to wait

[vlad20012]

bors r+

[bors]

Build failed:

• check

[Kobzol]

bors retry

EDIT: Oops, this is not my PR, I automatically retried it, since I got a CI bors notification in e-mail :D

[bors]

Build succeeded:

• check