Add cuda-cudart-static to generated conda environments

[marcinz]

Adding this package should resolve the build issues that have cropped up recently.

Issues that came up in the process:

• Fixed the runner that does not exist anymore
• Update cuda version to a format introduced in another PR
• Update the documentation
• Change to using labels for the nvidia channel
• Fix a test issue that was crashing the tests on OS X
• Do not use GPU runners for CPU tests
• Provide an option for not building the profiler and turn it off in the CI builds (provides a good speedup)
• Switch to using cuda 12.2.2
• Require that cuda is installed (__cuda requirement) in the meta.yaml, but do not request any specific version
• Add openmpi to requirements if we build with UCX

[marcinz]

This would also need to be updated in meta.yaml so that the metadata is tagged on the output conda package

Will do, but, for now, CI does not seem to be picking the jobs up.

Edit: actually, it seems to already be in meta.yaml.

[marcinz]

@manopapad Do we need to run mypy if it runs as a part of pre-commit? Currently, we are running it for every build, but that does not seem to add any value if I am not wrong.

Also, we seem to be getting some error related to multiprocessing in OS X tests. I wonder if you know what the problem is off the top of your head.

[manopapad]

Do we need to run mypy if it runs as a part of pre-commit?

Sorry, I missed this. I don't think it's necessary to run mypy again if it's getting run as part of pre-commit (assuming we have something like pre-commit-ci running pre-commit also on the server side; we can't guarantee that developers will have run it locally before committing).

[marcinz]

Yes, on this repo we have pre-commit.ci.

[marcinz]

Actually, I will just leave mypy jobs as they are for now since we are not getting too much development going on here atm.