refactor CUDA versions in dependencies.yaml

[jameslamb]

Description

Contributes to rapidsai/build-planning#7.

Proposes splitting the cuda-version dependency in dependencies.yaml out to its own thing, separate from the bits of the CUDA Toolkit this project needs.

Benefits of this change

• prevents accidental inclusion of multiple cuda-version version in environments
• reduces update effort (via enabling more use of globs like "12.*")
• improves the chance that errors like "conda recipe is missing a dependency" are caught in CI

[jameslamb]

1 build, conda-python-tests / tests (11.4.3, centos7, amd64, 3.9, v100, earliest), is failing with stuff like this

INTERNALERROR> Traceback (most recent call last):
INTERNALERROR>   File "/opt/conda/envs/test/lib/python3.9/site-packages/xdist/workermanage.py", line 386, in process_from_remote
INTERNALERROR>     warning_message = unserialize_warning_message(
...
INTERNALERROR>   File "/opt/conda/envs/test/lib/python3.9/asyncio/events.py", line 642, in get_event_loop
INTERNALERROR>     raise RuntimeError('There is no current event loop in thread %r.'
INTERNALERROR> RuntimeError: There is no current event loop in thread 'Dummy-8'.

(build link)

Not sure if it's relevant, but that is the only job in the matrix testing with an older driver.

I tried adding the CTK back to the environment for that job: 9ce9fc8.

Let's see if it helps.

[jameslamb]

From discussion offline with @AjayThorve ... I just pushed 84ce513 putting a ceiling on panel.

I did this in the bluntest way possible (applying it to all conda envs and pyproject.toml) just to test if it fixes the problem, but that might not be ideal. I'm especially skeptical of the pyproject.toml change... if that makes it out into wheels that users install, having a ceiling on panel is going to be disruptive.

If this works, I'm planning to remove that ceiling from pyproject.toml, do you agree @AjayThorve ?

[AjayThorve]

If this works, I'm planning to remove that ceiling from pyproject.toml, do you agree @AjayThorve ?

Yeah if this works, there haven't been any issues with other envs, and seems like the issue is running parallet pytests with updated panel is causing it. In that case, all we need to do is remove the current upper pin for panel from run dependencies and add a specific panel dependency with >=1.0,<1.3.4 to test_python section (under just conda)

[AjayThorve]

/merge