✅ Use Ruff rules to ensure safe lazy-loading of rich

[nathanjmcdougall]

To avoid regressions like #1294 occurring in the future, I propose a different way of lazy-loading rich using importlib.util.find_spec. I also propose using the TID Ruff rules to enforce the lazy-loading logic in the future.

This provides a framework to systematically apply other lazy-loading migrations safely.

[svlandeg]

Thanks for working on this! I'm putting this in draft as long as the CI fails - feel free to mark as ready for review when the CI goed green!

[nathanjmcdougall]

@svlandeg Alright, I think it's good now. I can't add labels so that check's failing, but everything else is green.

[svlandeg]

Great, so a recap.

This PR does a number of related things:

1. Avoid importing rich_utils at the top of any module, ensuring its lazy loading from #1128 (flake8-tidy-imports)
2. Avoid importing rich anywhere in the codebase, except in rich_utils.py and docs files (flake8-tidy-imports.banned-api)
3. Refactor how to define and store HAS_RICH: using importlib.util.find_spec("rich") instead of a try-catch import in core.py
4. Fetching the HAS_RICH var from core.py in both main.py and cli.py to avoid code duplication and have easier maintenance in the future
5. Refactor the tests to use HAS_RICH and pytest.MonkeyPatch instead of setting the typer.core.rich variable directly.

To me, 1 & 2 are the crucial points of this PR: they ensure we're not importing rich or rich_utils by accident. This would have caught #1290 as well as a similar issue we encountered in #828: it will be nice to set things up so that we can't hit ourselves with this again 🥲

3 follows out of this because you can't import rich anymore, and in fact having a HAS_RICH variable feels like a nicer solution as well, as compared to our old solution of rich either being an imported library, or None.

4 was added to this PR by my suggestion as I think it's more clean to define this functionality only once and then import.

5, the refactoring of the tests, needs to happen anyway when switching from the rich variable check for None to the boolean HAS_RICH. Using MonkeyPatch to do this feels more elegant to me than what we had before.

So, all in all, this looks (really) good to me. Thanks again @nathanjmcdougall for all your work on this!

I'm passing this on internally to Tiangolo's reviewing queue, he will update here once he's had a chance to review 🙏

[nathanjmcdougall]

After this PR is done, I plan do the same but with shellingham.

[github-actions]

📝 Docs preview for commit d8ffd89 at: https://07df5e78.typertiangolo.pages.dev

[tiangolo]

Awesome, looks great! Thanks for all the work on this @nathanjmcdougall! 🙌

And thanks @svlandeg for the thorough explanation of everything and the reasons for it all. 🚀