Install optional boltztrap2 and pygraphviz in CI

[DanielYang59]

Summary

• Install some optional dependencies in CI, to fix [Dev] Multiple dependencies for CI missing #3684.
• Fix patch for bader_caller (setting PATH to empty is easier than changing the patch location, and serves the same result where bader cannot be found in PATH)

Partially installed on Ubuntu

• boltztrap2: only with pip (uv wouldn't work for now Installation of BoltzTraP2 fails astral-sh/uv#5816 (comment))
• pygraphviz: installation guide

[janosh]

do we want to merge this as partially complete and leave some of the todos for later?

[DanielYang59]

Thanks for asking. I was facing some resistance while trying to install them (maybe that's the reason they're not installed in CI). And what make things worse is that I never use any of these.
Some help would be appreciated:

• graphviz seems to install just fine but is still skipped in CI for some reason (check the end of log for split 2).
• boltztrap2 is installed in Ubuntu, and revealed one broken unit test

UPDATE: boltztrap2 installation doesn't look quite reliable for some reason? It used to work just fine, but currently show some installation errors. We would have to drop it if so.

[DanielYang59]

@janosh Perhaps we could merge this as partially finished (as you suggested) and I would work towards the remaining parts shortly. The CI test failure is reported in #3969 and would be fixed by #3972, thanks!

[janosh]

thanks @DanielYang59! 👍

[DanielYang59]

Regarding the installation of BolzTraP2 on Windows, I'm afraid it wouldn't be an easy fix owing to the difference in cmake behavior. Let's install it on Linux only for now until we have a solution.

I would look into why uv failed to install it on Linux later.

[DanielYang59]

@janosh Forgot to mention this, I have to revert your update -> |= change here c67abbc because it seems to break unit test.

I haven't got a chance to look closer yet, perhaps the operator is not well defined for edges

[janosh]

yeah i noticed that, that was sloppy refactoring on my part. thanks for fixing!

perhaps the operator is not well defined for edges

that's very likely correct

[DanielYang59]

that was sloppy refactoring on my part.

pygraphviz was not installed in CI at that time, so there's no way for you to catch that :)

[janosh]

no, but i could have noticed that g.edges[u, v, k] is unlikely to be a dict in which case the first thought is to check whether the class in question implements __or__. (if not, |= is undefined behavior)