sinks, under wal: :none (the direct ack path) only: no sink failed its
last publish within listener.ready_sink_error_window_ms. Reason publish_failed.
In that mode the broker is the durable store and the provider is the retry, so
a node whose sinks refuse every publish answers every hook 503; unready tells
the load balancer to send them elsewhere. A sink that publishes again clears
its own entry (Ankusa.Health.SinkErrors), so the node is ready the moment the
broker is back. Under wal: :disk and :dual the queue retries a failing
sink, and the hook is acked regardless: there is nothing to report here.