ENH: Add M/EEG processing

[larsoner]

Before merging …

• Changelog has been updated (docs/source/changes.md)
• Add support for MEG+EEG data
• Reenable SSP for EEG
• Add Check that conditional that empty-room cov cannot be used for MEG+EEG data EDIT: Already exists
• CIs green
• Reports checked for proper sensor type plots (MEG and EEG) for each step (SSP, sensor plots/evoked, inverse)

@hoechenberger @agramfort sound like a plan? Also, why was SSP for EEG not supported / explicitly an error before? Looks like the error conditional that I removed here was added in #92 or so.

Closes #601

[larsoner]

Not quite ready yet, some plots missing from the report

[larsoner]

Ready for review/merge from my end.

Actually what I thought was missing was actually correct -- the empty-room plots should only have MEG channels. All other plots seem to reasonably contain MEG+EEG.

Also fixes a bug where STIM channels were included in decoding (whoops!). Now when we if cfg.analyze_channels; epocs.pick(cfg.analyze_channels) in a few places, we else: epochs.pick(cfg.ch_types) to restrict to the relevant data channels.

https://output.circle-artifacts.com/output/job/11d22b26-cfd0-4c25-a1f0-ea9197d5c3c0/artifacts/0/site/examples/ds000248/sub-01_task-audiovisual_report.html

[larsoner]

Okay there was still a bug with picking. I changed the default cfg.analyze_channels from 'all' which would include STIM channels to 'ch_types', which means "use cfg.ch_types".

I also fixed a doc bug where the "used in this example" was not determined properly because things like config['decode'] is only half the check, we also need config['conditions'] to be specified in order for decoding to be done.

[agramfort]

🎉 🙏