This is a similar issue to the one fixed via #8775

Currently, viz.plot_compare_evokeds()'s combine='gfp' argument makes it compute RMS, not GFP.

However, it should calculate GFP for EEG and RMS for MEG data. The combine='gfp' option should remain unchanged and simply pick the correct calculation depending on the channel type(s). This means that in case of M/EEG data, we'd have to apply different combine functions per subplot, so this is not a totally non-trivial change as far as I can tell (but might be wrong).