[prover] Handle connection between orchestrator and broker with multiple pending requests

[spalladino]

As @alexghr has identified, the orchestrator(s) may have 1000s of requests pending sent to the broker. If we use the default json-rpc abstraction, that means one awaited promise on an http connection per request, which will quickly exhaust the number of connections we can handle.

Instead, consider a long-lived connection like a websocket. The broker should have a stub on the orchestrator's side that handles this, and the interface to the orchestrator should be a simple one where it sends a request and eventually receives a callback. The broker should accept a job id for each request, and include it in the callback with the response.