Relax pinnings on several dependencies, remove nvidia channel

[bdice]

Description

This relaxes some pinnings on numba-cuda, numba, and pandas to ensure that the latest nightly of cudf can be used. I also removed the nvidia channel as it is no longer needed.

This also fixes cuopt_cli test which was failing due to missing pyenv

[gforsyth]

I think some of these changes also need to be set in the rattler recipes

[bdice]

Good call @gforsyth, normally I catch myself on that but not this time!

[gforsyth]

[rgsl888prabhu]

Thank you @bdice for the PR.

[bdice]

@rgsl888prabhu Could you take a quick peek at the install instructions? I just pushed some simplifications there while I was ripping out -c nvidia. We don't need to specify Python and CUDA minor versions in our install commands, it's best to let conda figure that out. It will work better with existing environments, and requires fewer updates on our side for every new CUDA/Python release.

[rgsl888prabhu]

@rgsl888prabhu Could you take a quick peek at the install instructions? I just pushed some simplifications there while I was ripping out -c nvidia. We don't need to specify Python and CUDA minor versions in our install commands, it's best to let conda figure that out. It will work better with existing environments, and requires fewer updates on our side for every new CUDA/Python release.

cuOpt packages are in nvidia channel, I think intsall instruction would require nvidia channel for cuOpt. https://anaconda.org/nvidia/repo

[bdice]

Sorry. Then my advice about removing nvidia only applies to RAPIDS and not cuOpt... I'll revert those changes.

[rgsl888prabhu]

@bdice I am working on fixing the error in tests, will push it to this PR and also revert changes to README

[rgsl888prabhu]

/merge