Peak memory usage measurements may be inaccurate or missing on Linux #161

Thank you for providing an independent benchmark of Python type checkers!

If I am understanding the code here correctly, you are using a (10 ms) polling strategy to measure peak RSS on Linux. This approach is prone to errors, since fast commands may not receive any measurements at all. And even for slow commands, the last 0 - 10 ms in peak RSS will be missing. Presumably, the end of the program is where most type checkers have their highest memory usage, so it would be good to include those last 10 ms.

I think you could use wait4 to get accurate peak RSS measurements on Linux, which would probably also result in a simplification of the code.

Disclaimer: I work on ty (but this issue is not about something that would affect ty in a specific way)