pick some real Python projects and attempt to port them to uv with universal locking

[BurntSushi]

I'd like to start small (as in, a small dependency tree) but with a project that benefits from having a platform independent lock file. So probably a project that utilizes the universal locking of Poetry or PDM.

I'd also like to try locking packse's pyproject.toml and see what happens.

The idea is to see what issues arise with our universal locking implementation in real projects.

Ref #3347

[mjclarke94]

Pretty much anything which leverages torch would be a good stress test given you need to use different index urls on a per platform basis.

[adeebshihadeh]

We're excited to switch for openpilot.

Previous attempt here: commaai/openpilot#32530.

[helderco]

Curious to try it in dagger modules, but I'll need to catch up on these changes.

[jab]

bidict has a pretty small dependency tree (no runtime deps, and just a handful of test deps), in case it's a good candidate for testing this with -- it would exercise resolving different dependencies for different versions of python as well as for different dependency groups all into the same lockfile. It currently uses uv pip compile-generated requirements files (rather than Poetry or PDM). Looking forward to simplifying all those test.txt requirements files (under the dev-deps/* subtree) that have been required without this feature.

[zanieb]

I ported packse in astral-sh/packse#183 if anyone is interested in what that looked like.