Take the information which stubs to exclude from the METADATA.toml files. On one hand this is necessary for #10722, i.e. to correctly handle requires_python, on the other hand this allows us to move a "stub incomplete" marker into the METADATA.toml file, and use it for other purposes than just the pyright tests. E.g. instead of relying solely on pyright to prevent regressions of "complete" stubs (which doesn't work great, since pyright doesn't understand Incomplete), we could have a separate script in CI.