[ci] speed up conda setup for some jobs

[jameslamb]

In #5648 , I saw that conda create-ing an environment in one shot was significantly faster in some situations than conda create-ing an environment and then updating it with conda install or conda update (#5647 (comment)).

That PR was for Windows CI jobs.

This PR proposes similar changes to some of the macOS and Linux jobs.

• moves lint, if-else, and swig job code up earlier in .ci/test.sh
  • swig job doesn't need a conda environment, so time in that job spent waiting for conda create is wasted time
  • lint and if-else have very different conda-installed dependencies than the other jobs, so they can be totally self-contained and just conda create their own environment with everything they need

benefits of this change

I observed the following timing differences in these jobs on this PR, compared to master:

job	latest master	this PR
(GHA) macOS if-else	06m33s	07m04s
(GHA) check-docs	04m08s	03m30s
(Azure) Linux inference	04m52s	04m36s
(Azure) Linux swig	04m55s	04m02s
(Azure) Linux_latest inference	04m07s	03m56s
(Azure) macOS swig	07m42s	05m05s

this PR:

• macOS GitHub Actions: https://github.com/microsoft/LightGBM/actions/runs/3870849816
• static_analysis GitHub Actions: https://github.com/microsoft/LightGBM/actions/runs/3870849804
• Azure: https://dev.azure.com/lightgbm-ci/lightgbm-ci/_build/results?buildId=14120&view=results

latest master build:

• macOS GitHub Actions: https://github.com/microsoft/LightGBM/actions/runs/3849602797
• static_analysis GitHub Actions: https://github.com/microsoft/LightGBM/actions/runs/3849602798
• Azure: https://dev.azure.com/lightgbm-ci/lightgbm-ci/_build/results?buildId=14109&view=results

Notes for Reviewers

Given all the things that can change from run-to-run, it's hard to trust small differences in the timings based on 1 or even a few of runs. But I'm still pretty confident that these changes make these CI jobs at least a small amount faster, without reducing their stability.

If this PR is accepted, I'll propose another one after that consolidates these into a single conda create for the remaining tasks:

LightGBM/.ci/test.sh

Line 37 in 61e464b

conda create -q -y -n $CONDA_ENV "python=$PYTHON_VERSION[build=*cpython]"

LightGBM/.ci/test.sh

Lines 126 to 140 in 61e464b

	# re-including python=version[build=*cpython] to ensure that conda doesn't fall back to pypy
	conda install -q -y -n $CONDA_ENV \
	cloudpickle \
	dask-core \
	distributed \
	joblib \
	matplotlib \
	numpy \
	pandas \
	psutil \
	pytest \
	"python=$PYTHON_VERSION[build=*cpython]" \
	python-graphviz \
	scikit-learn \
	scipy || exit -1

[jameslamb]

@jmoralez could you help with a review on this? I think it should be quick and non-controversial.

[jameslamb]

thanks @jmoralez !

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed.
To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues
including a reference to this.