Endless client job stream registration

[npepinpe]

Describe the bug

When registering a new client job stream from the gateway to any broker, the brokers are successfully registering the streams, but not responding to the gateway, causing the gateway to retry over and over.

While this still allows pushing, it creates a lot of noise and the impression that the streams are not successfully registered.

The underlying cause is the usage of the ClusterCommunicationService#subscribe(String, Function<byte[], M>, BiConsumer<MemberId, M>, Executor executor) - any subscriber which is a consumer will never send a response back. However, the client in this case expects a response.

Expected behavior

Client streams are properly registered on both sides.

[npepinpe]

One hurdle here is there's no easy way to introspect the registered client streams. So our tests were passing because we just check if the server has registered the stream (which it has), but we have no real way to catch that the client is in an endless loop.