When retrying internal requests, backoff only once for the full batch of retries

We currently wait for each internal retry request (TOLEADER or FROMLEADER requests). This can cause a long wait time when retrying many requests and can timeout the client. We should instead wait once and retry the full batch of errors.