Improve the efficiency of trace stitching

From the way benchmarks have been responding to recent PRs, it certainly seems like side exits could be made more efficient, especially in the happy path. This is especially true of _DYNAMIC_EXIT, which needs to do quite a bit of work for a successful stitch... perhaps the best option here is to just reduce the number of _DYNAMIC_EXITs that we run.

This could also include recompiling stitched traces together into the same executor to essentially remove stitching overhead, too.

Linked PRs

• gh-122295

该 issue 没有指定文件或测试。首先检查 benchmark 响应以及 _DYNAMIC_EXIT 附近的 trace-stitching 路径；确定减少 dynamic exits 或重新编译 stitched traces 是否可行。当能够证明相关 benchmark 上的 stitching 开销降低时，即视为完成。