project-root inference in your code-health workflow

Hi! I'm the author of [pyscn](https://github.com/ludo-technologies/pyscn). I came across this repo and noticed you're running it in your weekly code-health workflow. Thanks for using it.

I also read the comment in `code-health.yml` about pyscn inferring the wrong project root and silently resolving only part of the imports, which inflates the score. That's a real problem on my side and I'd like to fix it rather than have you work around it. If you remember what the wrong root looked like (or a run where the sanity check fired), I'd appreciate the details. Same for having to run `analyze` twice to get both JSON and HTML; that should be a single invocation.

Separately, there's a GitHub App that runs the same analysis on a schedule and posts the reports for you, in case it saves some upkeep:
https://github.com/apps/polyscan-app

If anything else comes up, feel free to let me know.

检查 code-health.yml 中的注释和当前调用，然后收集错误的项目根目录详细信息，或收集一次触发 sanity check 的运行记录。要求的结果是让 pyscn 正确解析完整项目，并通过一次 analyze 调用同时生成 JSON 和 HTML，但尚未确定实现位置和范围。

由索引模型根据 Issue 内容生成。