Allow making evokeds per event type

Describe the new feature or enhancement

It is very common that you want Evokeds for each event type, for example:

epochs = mne.Epochs(raw, events, {'left': 1, 'right': 2})
evokeds = [epochs['left'].average(), epochs['right'].average()]  # <-- Make this easier

Describe your proposed implementation

Add a parameter to the Epochs.average method:

evokeds = epochs.average(per_event_type=True)

Furthermore, the .comment attribute should be set to the event type.

Describe possible alternatives

Keep the following list comprehension as canonical solution:

evokeds = [epochs[cond] for cond in epochs.event_id.keys()]
for ev, cond in zip(evokeds, epochs.event_id.keys()): 
    ev.comment = cond

[drammock]

summarizing a conversation with @agramfort during office hours:

1. Let’s not bother supporting the list-of-lists and stick with just the 2 options for now.
2. don’t return a dict, return a list (because of how we do write_evokeds)
3. make sure to set the evoked.comment to the value of the event_id key (so it’s easy to recover )
4. then all that is left is we just need to decide the right name for the parameter (I’m happy with grouping too)

[agramfort]

works for me. Sleeping over it I like

epochs.average(by_event_id=True | False)