fix: #6558 don't try to repair invalid self-imports.

[MarcusGrass]

Current attempts break idempotence.

As the tests show, paths ending with self are now untouched, this PR just removes some code (and adds/modifies some tests).

A potential change here would be to keep this part of the code:

UseSegmentKind::Slf(None) if self.path.is_empty() && self.visibility.is_some() => {
    self.path = vec![];
    return self;
}

In that case

use self;

would be removed, but not:

use self::self;

Which woud also fix the idempotence issue, but formatting away use self;. Although maybe it's best that the compiler inform the user that they're doing something wrong instead of 'sweeping it under the rug' so to say.

[ytmimi]

I'd still like to normalize foo::self -> foo. Maybe we can special case this to not normalize self::self -> self to prevent idempotent issues in this really obscure case.

[ytmimi]

I take it this is where we get rid of use self;? I don't think there's any harm in keeping this behavior as long as we handle use self::self in a way that's idempotent.