Odd ecosystem results on typeshed

[AlexWaygood]

Seems like there's something up with the ecosystem check when it runs on typeshed currently. We've had issues on several PRs where the ecosystem check has produced surprising results that don't make logical sense and can't be reproduced locally:

• [flake8-pyi] Allow simple assignments to None in enum class scopes (PYI026) #11128 (comment)
• [flake8-pyi] Implement PYI062 (duplicate-literal-member) #11269 (comment)
• [pyflakes] Allow forward references in class bases in stub files (F821) #10779 (comment)

No idea what's up here

[MichaReiser]

Would you mind listing some of the "odd" examples and explaining why they're odd?

[AlexWaygood]

Would you mind listing some of the "odd" examples and explaining why they're odd?

In all three comments I linked to above, the ecosystem check either surfaced new lint errors that could not be reproduced locally when running ruff with the PR branch, and/or failed to report errors going away even though locally running ruff with the PR branch led to existing lint errors going away