ci: Add a linkerd install workflow

[olix0r]

We don't currently test that proxy changes are at least usable in a
basic Linkerd installation. This adds a CI workflow to test all Linkerd
code changes against a real k8s cluster with the control plane from the
most recent edge release.

[hawkw]

looks like the build is busted on the latest nightly --- this looks like a regression to an issue i feel like we've seen before, at first glance

[mateiidavid]

Would it be okay to remove this? Thinking it might be redundant if we have it as a dependency for just linkerd-install?

linkerd-install *args='': _tag-set k3d-load-linkerd _linkerd-crds-install && _linkerd-ready

[hawkw]

looks like the build is busted on the latest nightly --- this looks like a regression to an issue i feel like we've seen before, at first glance

Oh, apparently this is slightly different, it's a warning now. Should be trivial to fix.

[hawkw]

Hmm, the warning (``) is actually somewhat less trivial to fix, as we can't just turn types like

            impl svc::Service<
                    http::Request<http::BoxBody>,
                    Response = http::Response<http::BoxBody>,
                    Error = Error,
                    Future =  impl Send,
                > + Clone,

into

            impl svc::Service<
                    http::Request<http::BoxBody>,
                    Response = http::Response<http::BoxBody>,
                    Error = Error,
                    Future = impl Future<Output = Result<http::Response<http::BoxBody>, Error>> + Send,
                > + Clone,

when we want to use impl Send to indicate an additional Send bound in addition to the Future bound, as this results in an error like

error[E0277]: `impl std::marker::Send` is not a future
  --> linkerd/app/outbound/src/http/logical.rs:97:30
   |
97 |                     Future = impl Future<Output = Result<http::Response<http::BoxBody>, Error>> + Send,
   |                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ `impl std::marker::Send` is not a future
   |
   = help: the trait `futures::Future` is not implemented for `impl std::marker::Send`
   = note: impl std::marker::Send must be a future or must implement `IntoFuture` to be awaited
   = note: required for `stack::map_err::ResponseFuture<(), impl std::marker::Send>` to implement `futures::Future`

For more information about this error, try `rustc --explain E0277`.

which...feels incorrect to me.

We will probably want to change our warning configuration to allow the warning opaque_hidden_inferred_bound, and probably consider opening a Rust issue to report that this warning prevents us from writing code that uses opaque types in associated types?

[hawkw]

#2275 should hopefully fix the nightly build failure.

[hawkw]

very nice, the new just recipes will also make manual testing with dev builds of the proxy much easier!