MRG, MAINT: Lower memory again

[larsoner]

Another round. This time I cut down plot_mne_dspm_source_localization to deduplicate its content with other examples/tutorials, and removed a bunch of verbose='error's that we no longer need.

[agramfort]

@drammock merge if you're happy. it touches your tutorial text

[larsoner]

Okay so it looks like it's not specific examples that are the problem. If I build with:

BUILD_DEV_HTML=1 mprof run --python sphinx -b html -nWT --keep-going -d _build/doctrees . _build/html

I get:

I'll try tacking on an xvfb-run (which uses MESA GL, just a never version than CircleCI), but this ever-increasing memory usage is not good.

[larsoner]

Okay I think I know the culprit. If I run using the pattern plot_m, I get this:

You can see a bunch of deltas (good), but also a step at around 120 seconds (bad). The step is the problem. If I run plot_montage, I get:

Compare this to one of our "large memory" examples, plot_mne_dspm:

In other words, the plot_montage example allocates some memory that never gets freed. It creates probably 30 PyVista plots, and for some reason these must never get GC'ed properly. I'll look into it.

In the meantime I've pushed a commit here to stop building it by removing the plot_ from the name, and we'll see if the [circle full] is green. If it is, I'll merge, and we'll see if master is green. Then hopefully tomorrow I can track down what's causing the memory to linger so that we can restore the plot_. But we can also do this sometime after release, so at least CircleCI will no longer be stopping us -- hopefully we can release tomorrow!

I'll also try modifying sphinx-gallery to capture this residual memory problem...

[agramfort]

success ! https://app.circleci.com/pipelines/github/mne-tools/mne-python/3745/workflows/6bb7974f-e742-4249-815e-ce37558c16e6 thanks so much @larsoner !

…