rpc半同步模式访问, 自定义retry policy, 偶发FATAL日志, A previous non-backed-up call responded

**Describe the bug (描述bug)**

**To Reproduce (复现方法)**
半同步访问 + 自定义的retry_policy，client发压，很容易复现, 服务不会挂掉。
FATAL日志:controller.cpp:609] Check failed: current_id() == info.id (9925669421080 vs 9925669421077). error_code=1008
controller.cpp:844] Check failed: false. A previous non-backed-up call responded

**Expected behavior (期望行为)**
rpc半同步模式访问 + 自定义retry policy, 不出现偶发FATAL日志，是否在使用retry policy时候需要注意什么参数。

**Versions (各种版本)**
OS:
Compiler:
brpc:
protobuf:

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

Start by inspecting controller.cpp around lines 609 and 844, then reproduce the reported failure with semisynchronous RPC access and a custom retry policy. Check the available issue context and retry-policy parameters while investigating the FATAL logs; done means the scenario no longer emits these failures.

Written by the indexing model from the issue text.