BLD: separate lightpath and lightpath[gui] subpackages

[tangkong]

Description

Separates lightpath base and gui-related dependencies

• pip: lightpath and lightpath[gui]
• conda: lightpath-base and lightpath (includes gui) I could not get conda builds working, we leave this split to the conda feedstock

Motivation and Context

Lightpath keeps dragging qt dependencies along with it

How Has This Been Tested?

pip installs locally. Conda eventually, but let's see if CI takes

Where Has This Been Documented?

This PR

[ZLLentz]

Suggested change

	script: python -m pip install . -vv
	script: {{ PYTHON }} -m pip install . -vv

I'll look for the docs on this

[ZLLentz]

https://docs.conda.io/projects/conda-build/en/latest/user-guide/environment-variables.html

[tangkong]

https://github.com/pcdshub/lightpath/actions/runs/10564732964/job/29267769359#step:12:49

Attempting to finalize metadata for lightpath-base
Error: Failed to render jinja template in /home/runner/work/lightpath/lightpath/conda-recipe/meta.yaml:
'PYTHON' is undefined

[ZLLentz]

https://github.com/pcdshub/pcdsdevices/blob/b33b7c4d6a372cb877458f7e446c90d0d083518f/conda-recipe/meta.yaml#L15

[tangkong]

Maybe this was the warning shot that suggests that something deeper was wrong with the conda build process

[ZLLentz]

I see now that the recipes that do the *-base thing don't use {{ PYTHON }}, my bad for the red herring

[tangkong]

After quite a bit of squinting at conda build logs, I've arrived at a point where the conda build step succeeds, but the test environment gets clobbered/lost and cannot be used for the rest of the test suite. This is despite the environment artifact being created and uploaded sensibly

The changes to the pip-side of things have worked from the onset, and I think are good moving forward. A nearly identical recipe works on the lightpath-feedstock and passes their build tests, so I think we'll revert this repo's recipe to the combined recipe and only split up the feedstock recipe.

[tangkong]

adding --no-deps managed to remove the clobber warnings I think. There definitely was some double installations going on, I think what happens is that without --no-deps, the pip install installs all the package dependencies, then conda tries to follow the recipe and install them again.

We still lose the test_env folder though 😢

[ZLLentz]

The remaining py311 failures come from pinning pytest<7.2.0, before this version it wasn't marked as noarch so the allowed versions are very pinned down

[tangkong]

Ah we unpinned this a while ago. This was from an era where pytest-qt wasn't compatible

[ZLLentz]

It's all GREEEN

[ZLLentz]

This is a great split that will help a lot in various cases!