Add a benchmark that heavily uses type annotations

This is a feature request.

Would it make sense to add a benchmark for code that heavily uses annotations, generics, and other typing features? I even have a suggestion for such benchmark: running mypy on itself (on its own code). The run currently takes around 10-20 sec. (depending on machine). If this is too long, other option would be to run mypy on one of typeshed stubs, this will take less than a second.