能否将rq视作一个unbounded队列，来对请求进行缓存处理

**Is your feature request related to a problem? (你需要的功能是否与某个问题有关?)**
代码如下，业务需求是：rpc提前返回给上游，然后server读执行后续的耗时操作（落盘）。

**Describe the solution you'd like (描述你期望的解决方法)**
1、我能否将brpc的rq视作一个mpmc的队列，这样就不用在回调中入队，后台消费线程出队？
2、如果可以，能否在done->Run()之后，“用户的耗时操作”之前，利用令牌桶进行控速

**Describe alternatives you've considered (描述你想到的折衷方案)**

**Additional context/screenshots (更多上下文/截图)**

The issue names brpc's rq, the done->Run() callback, and a token-bucket limiter, but no files or tests. Start by tracing rq and the RPC callback lifecycle to determine whether rq provides the requested MPMC and unbounded-queue behavior. Done should be a documented design decision, with implementation and tests only if the project accepts the proposed request-processing model.

Written by the indexing model from the issue text.