Is the plan to only rely on retries/hedging for avoiding slow backends?
If there is issues between a client and backend that is only visible to the client we, currently, can't act on that based on the ClientStats nor the server LoadBalancingFeedback?