builder,util: add convenience methods for boxing services

[davidpdrsn]

This adds a couple of new methods to ServiceBuilder and ServiceExt:

• ServiceBuilder::boxed
• ServiceExt::boxed
• ServiceBuilder::clone_boxed
• ServiceExt::clone_boxed

[hawkw]

I commented on some docs nits and stuff, but I'm mostly good with this. It would be nice to fix the wrong syntax in links before merging, though.

[hawkw]

yeesh...i feel like it's almost worth introducing a named Layer type for BoxService just to avoid having this type signature...but, I think that might be a breaking change...

[davidpdrsn]

I went down that route initially actually but you end up having to have the request type as a type param on the layer:

// have to have `T` here otherwise its unconstrained in the layer impl
pub struct CloneBoxServiceLayer<T> {
    _request: PhantomData<T>,
}

impl<S, T> Layer<S> for CloneBoxServiceLayer<T>
where
    S: Service<T> + Clone + Send + 'static,
    S::Future: Send + 'static,
{
    type Service = CloneBoxService<T, S::Response, S::Error>;

    fn layer(&self, inner: S) -> Self::Service {
        CloneBoxService::new(inner)
    }
}

That could be workable to leads to inference issues when actually using the builder method:

---- src/builder/mod.rs - builder::ServiceBuilder<L>::clone_boxed (line 754) stdout ----
error[E0284]: type annotations needed
  --> src/builder/mod.rs:765:6
   |
14 |     .clone_boxed()
   |      ^^^^^^^^^^^ cannot infer type for type parameter `S`
   |
   = note: cannot satisfy `<_ as Service<Request>>::Future == _`

[hawkw]

ugh, never mind. it's probably better to avoid introducing phantomdata anyway, as they will result in rustc having to typecheck a larger type, potentially impacting build time... this is fine, then.

[hawkw]

TIOLI but I might have called this boxed_clone? although on the other hand, this isn't consistent with the name of the return type...idk.

[davidpdrsn]

Right I went with this name because it matches the service name, which we can rename since its not released. CloneBoxService or BoxCloneService seem kinda same same to me though.

[hawkw]

nit, take it or leave it: maybe worth adding a note comparing this with box_service, like I mentioned on the ServiceExt methods? not a blocker...

[hawkw]

gosh... since the CloneBoxService::layer() method hasn't been released yet, we could potentially replace the LayerFn with a named Layer type. IDK if it's worth it just to get rid of this type signature, but...wow.

[hawkw]

ugh, never mind. it's probably better to avoid introducing phantomdata anyway, as they will result in rustc having to typecheck a larger type, potentially impacting build time... this is fine, then.