ci: split windows x86 job

[kmantel]

Windows x86 ci runs can fail because the container runs out of memory
during testing. This is caused by memory leaks in psyneulink and limited
available memory on the containers. As a workaround, running the full
set of tests over multiple jobs reduces peak memory usage.

Three jobs replace the original Windows x86 build, running pytest with:

1. -m llvm
2. -m "not llvm and composition"
3. -m "not llvm and not composition"

Jobs with the same OS, python version, and architecture overwrite
uploaded dist packages.

[github-actions]

This PR causes the following changes to the html docs (ubuntu-latest-3.11-x64):

No differences!

...

See CI logs for the full diff.

[jvesely]

Does splitting by tag also address the issue?
e.g.

          - python-version: '3.8'
            python-architecture: 'x86'
            os: windows
            extra-args: '-m llvm'

          - python-version: '3.8'
            python-architecture: 'x86'
            os: windows
            extra-args: '-m not llvm'

(splitting by other tags should work too)

[kmantel]

It's a good suggestion - running multiple pytest commands in the job immediately fails the entire job if one command fails, so the entire suite may not be run.

Is there an additional benefit to splitting by llvm? not llvm still gives the memory error in that case, so I'd need to make three jobs.

extra-args: '-m llvm'
extra-args: '-m not llvm and composition'
extra-args: '-m not llvm and not composition'

but just splitting by composition seems to be ok also

[github-actions]

This PR causes the following changes to the html docs (ubuntu-latest-3.11-x64):

No differences!

...

See CI logs for the full diff.

[jvesely]

Can you add a comment explaining the not allowed characters in artifact names?
Using >> $GITHUB_OUTPUT and steps.chosen-step-id.output below would be more targetted and avoid setting an env var for all follow up steps

[kmantel]

I will. It looks like only double quote is actually disallowed, so I can remove the other substitutions. But I also think space->underscore is nice although unnecessary

[jvesely]

does:

overwrite: matrix.python-architecture == 'x86'

work?
the dist packages should be identical and it would be a bit easier to read

can you also add one sentence about the necessary adjustments to upload to the git commit description

[kmantel]

It looks like overwrite won't allow expressions, only true/false, unless I can't figure how to do it correctly. overwrite: true can still work and would currently only affect the x86. If we needed to add other jobs later without overwriting, we'd have to change the name anyway

[github-actions]

This PR causes the following changes to the html docs (ubuntu-latest-3.11-x64):

diff -r docs-base/Mechanism.html docs-head/Mechanism.html
379c379
< <strong>comparison_operation</strong> argument, that is used to set the LinearCombination function’s <a class="reference internal" href="CombinationFunctions.html#psyneulink.core.components.functions.combinationfunctions.CombineMeans.operation" title="psyneulink.core.components.functions.combinationfunctions.CombineMeans.operation"><code class="xref any py py-attr docutils literal notranslate"><span class="pre">operation</span></code></a> parameter.</p>
---
> <strong>comparison_operation</strong> argument, that is used to set the LinearCombination function’s <a class="reference internal" href="IntegratorFunctions.html#psyneulink.core.components.functions.stateful.integratorfunctions.DualAdaptiveIntegrator.operation" title="psyneulink.core.components.functions.stateful.integratorfunctions.DualAdaptiveIntegrator.operation"><code class="xref any py py-attr docutils literal notranslate"><span class="pre">operation</span></code></a> parameter.</p>
diff -r docs-base/RefactoredLearningGuide.html docs-head/RefactoredLearningGuide.html
282c282
< <p>Notice that we no longer have to extract the target node from the <code class="xref any docutils literal notranslate"><span class="pre">add_backpropagation_learning_pathway</span> <span class="pre">method</span></code>, and can instead pass the targets as output nodes mapped to values, in a new parameter called <code class="xref any docutils literal notranslate"><span class="pre">targets</span></code> in the <a class="reference internal" href="AutodiffComposition.html#psyneulink.library.compositions.autodiffcomposition.AutodiffComposition.learn" title="psyneulink.library.compositions.autodiffcomposition.AutodiffComposition.learn"><code class="xref any py py-meth docutils literal notranslate"><span class="pre">learn</span></code></a> method.</p>
---
> <p>Notice that we no longer have to extract the target node from the <code class="xref any docutils literal notranslate"><span class="pre">add_backpropagation_learning_pathway</span> <span class="pre">method</span></code>, and can instead pass the targets as output nodes mapped to values, in a new parameter called <code class="xref any docutils literal notranslate"><span class="pre">targets</span></code> in the <a class="reference internal" href="EMComposition.html#psyneulink.library.compositions.emcomposition.EMComposition.learn" title="psyneulink.library.compositions.emcomposition.EMComposition.learn"><code class="xref any py py-meth docutils literal notranslate"><span class="pre">learn</span></code></a> method.</p>
287c287
< its constructor to be runtime parameters of its <a class="reference internal" href="AutodiffComposition.html#psyneulink.library.compositions.autodiffcomposition.AutodiffComposition.learn" title="psyneulink.library.compositions.autodiffcomposition.AutodiffComposition.learn"><code class="xref any py py-meth docutils literal notranslate"><span class="pre">learn</span></code></a> method.</p>
---
> its constructor to be runtime parameters of its <a class="reference internal" href="EMComposition.html#psyneulink.library.compositions.emcomposition.EMComposition.learn" title="psyneulink.library.compositions.emcomposition.EMComposition.learn"><code class="xref any py py-meth docutils literal notranslate"><span class="pre">learn</span></code></a> method.</p>

...

See CI logs for the full diff.