'complete' option in PublicationCache, refactoring

[milyin]

• added 'complete' option to query of PublicationCache
• refactored access to SessionRef object: added mthods for creating publisher, queryable, etc to it. This allowed to access the SessionRef without code duplication just to satisfy borrow checker
• Simplified SessionExt trait

Closes 'rust' checkbox of #668

[eclipse-zenoh-bot]

@{login} If this pull request contains a bugfix or a new feature, then please consider using Closes #ISSUE-NUMBER syntax to link it to an issue.

[OlivierHecart]

This broke The usage of SessionExt with Arc<Session>.

#[async_std::main]
async fn main() {
    use zenoh::prelude::r#async::*;
    use zenoh_ext::SessionExt;

    let session = zenoh::open(Config::default())
        .res()
        .await
        .unwrap()
        .into_arc();

    let publication_cache = session
        .declare_publication_cache("key/expression")
        .res()
        .await
        .unwrap();

    async_std::task::spawn(async move {
        publication_cache.key_expr();
    });
}

[milyin]

@OlivierHecart Fixed missing implementation of SessionExt for Arc<Session>.
Also modified SessionDeclaration trait: now this trait is the only source of declaration functions. Scheme is exaclty the same as in SessionExt
@p-avital could you please look at it too? I hope this approach to using lifetimes is ok, but it's worth to check

[p-avital]

This is actually a sneaky breaking change: the SessionDeclarations trait will have to be in scope for the methods to be called on Session, whereas this is currently unnecessary.

While I agree this makes a lot of sense, we should probably have a quick internal chat about whether or not we want to make that break (we have other more localized breaking changes in the next release, so I'm open to it)

[p-avital]

The lifetime handling looks correct to me (GATs could have been another way to reach the solution, but this pattern also allows some nice variance tricks if we ever need them).

There's a sneaky breaking change by moving some of Session's method to a trait. I think moving to the trait makes a lot of sense, but we might also want to re-add the intrinsic methods to avoid the breaking change.

This review will be updated to approved when a decision has been made regarding this breaking change.

[milyin]

Michael Ilyin — Today at 11:04 AM
Thank you! Speaking on breaking change - I don't mind to repeat declaration functions in impl Session. I have only one argument against it:

If even core API, like declare_publisher is implemented in a trait outside the Session, this makes SessionExt extension first class citizen. It's now not a some hack to add new methods to Session, but completely valid and supported way to extend it.

Pierre 'Fox' Avital — Today at 11:08 AM
I agree, in general, I'm always in favour of traitifying, although that does mean more effort is necessary on docs to help the user navigate the API, as trait implementations are generally less obvious than methods

[milyin]

Talked with @Mallets - the only concern from his point of view is navigation on documentation. Created task to not forget to take care of it #710. Ohterwise he also thinks that this update is ok. Will integrate it today