Create an in-memory implementation for the WorkerChannel

[gabotechs]

With the WorkerChannel abstraction shipped in #512, we can have a fully in-memory implementation that communicates any WorkerChannel client call with a local Worker.

The implementation will likely be small and pretty much a proxy to the Worker methods.

About wiring this up with the codebase, the in-memory comms should likely interact well with the present LocalWorkerContext in the TaskContext at the moment of calling ChannelResolver::get_worker_client_for_url:

• If LocalWorkerContext.self_url == url: fallback to the in-memory implementation of the WorkerChannel
• Otherwise, use whatever other WorkerChannel implementation is registered

This probably means that the ChannelResolver::get_worker_client_for_url should be extended with another task_ctx: &Arc<TaskContext> argument so that we can retrieve the WorkerChannel implementation from there.

One idea that comes to mind, is to just have a helper function that is used in the places where WorkerChannels are retrieved for specific URLs, and apply this logic of returning the in-memory implementation instead of the real one if LocalWorkerContext.self_url == url:

• datafusion-distributed/src/worker/worker_connection_pool.rs

  Line 130 in e3ac0fb

  let mut client = ch_resolver.get_worker_client_for_url(&target_url).await?;

• datafusion-distributed/src/coordinator/query_coordinator.rs

  Line 199 in e3ac0fb

  let mut client = channel_resolver.get_worker_client_for_url(&url).await?;