reduce the number of times of marshall for response

## Bug Report

Previously, server will reply a response to client if it receives a request, which will marshall the corresponding response. For a write request from client, all of primary and secondaries will response for it, which means the response's marshall operation is excuted by three times. We can do some optimizations for this situation, and don't marshall response for secondaries.

Given the marshall is heavy cost in cpu and mem, it will improve the throughput of our cluster if we implement these optimizations.

The issue names no files, tests, or entry points. Trace response handling for write requests across primary and secondary replicas, then verify that secondary responses no longer perform marshalling while client behavior and cluster throughput remain correct.

Written by the indexing model from the issue text.