RFC: in search of the perfect sample data

After revising lots of tutorials, I've formed some opinions about what the ideal sample dataset would look like. Here are some thoughts; please feel free to edit / add to the list; maybe if we reach some consensus we can find a dataset that meets the criteria and could replace sample dataset in most/all of the tutorials.

• probably should be neuromag and include fine cal, crosstalk, and a cHPI .pos file (current sample data lacks .pos)
• has mag, grad, EEG, EOG, and ECG (current sample data lacks ECG)
• has multiple subjects (useful for demoing corrmap, stats, ...)
• multiple runs per subject maybe? (for things like concatenating raws)
• a single Raw recording is ~90 seconds max, and/or has a low-ish acquisition sampling frequency (so we're not constantly cropping to save memory)
• raw file does not already contain projectors; separate empty-room file included
• I think the task (L/R aud/vis + catch trials) is pretty good as a demo, but might be nice to have reaction time data included (for things like the epochs image overlay). I'm certainly open to other ideas about the paradigm though.
• EEG channel names follow 10-20 / 10-05 naming conventions

Review the current sample dataset and the tutorial uses referenced in the issue, then compare them with the listed acquisition, channel, subject, run, and file requirements. Done means reaching consensus on a replacement dataset and its criteria; the issue does not name implementation files or tests.