Limit outstanding client requests to the cluster

It is currently possible for clients to accumulate large numbers of outstanding requests. In some situations, such as network thread saturation or disconnection from the cluster, it can happen that many requests queue up and then get submitted to the cluster very rapidly, overwhelming server processes.

To prevent this, we could institute a limiting mechanism to the number of requests that are outstanding on a client. This could be used to fail requests that exceed our limits without them ever being sent.

Start by tracing the client-side path that queues and submits outstanding requests to the cluster. Define how the limit is configured and how excess requests fail before being sent; done means clients cannot accumulate or rapidly submit requests beyond the chosen limit.

Written by the indexing model from the issue text.