Avoid --find-links in wheel jobs

[jameslamb]

Description

Contributes to rapidsai/build-planning#69
Contributes to rapidsai/build-planning#33

Proposes a stricter pattern for passing a pylibcugraph wheel from the wheel-build-cpp job that produced it into the wheel-build-python job wanting to use it (as a build dependency of cugraph). This change improves the likelihood that issues with the pylibcugraph wheels will be caught in CI.

Notes for Reviewers

See rapidsai/rmm#1586 and rapidsai/build-planning#69 (comment) for details on how I tested this approach.

[msarahan]

It's counterintuitive to me how this works, but I also think that it is a good idea. The counterintuitive part is how we're using constraints to specify the path to the file. It's not wrong, it just isn't a use for the constraints file that I had considered.

[jameslamb]

It's counterintuitive to me how this works, but I also think that it is a good idea. The counterintuitive part is how we're using constraints to specify the path to the file. It's not wrong, it just isn't a use for the constraints file that I had considered.

Thanks for the careful review @msarahan .

I agree, it is definitely counterintuitive at first (I tried several other ideas before landing on this one). I think because we're in this situation where we both want to use build isolation but also want very fine-grained control over some characteristics of the build environment. If we wanted much more control than this, it'd probably be time to just abandon build isolation (I mentioned that near the end of rapidsai/build-planning#79, a potential follow-up to this).

[jameslamb]

/merge