ENH: Remaining stuff for Epochs annotations

From https://github.com/mne-tools/mne-python/pull/9969, @adam2392 we should take care of these before 1.0. Creating this issue so that we don’t forget:

• Annotations will be lost on save. (#10019)
• It is not yet possible to add annotations to the Epochs object programmatically (via code)

The first one here seems less critical to me, so I think we could leave it for 1.1, and the second one is too hard for 1.0:

• Concatenating :class:~mne.Epochs objects that contain annotations is not supported, and any annotations will be dropped when concatenating.
• It is not yet possible to add annotations to the Epochs object interactively (through the plot window)

[larsoner]

I see that makes sense. Then set_annotations_by_epoch would set one “annotation” with the duration of the epoch by default. Alternatively, it can set multiple annotations with no duration.

Now we’re talking API… I think maybe a user could pass a list of length n_epochs with onset/duration/desc tuples inside or something. But we can hash that out based on a use case…

I am down to work on this down the line, but am quite busy recently. Maybe a newcomer developer would like to take this on?

… so I think we should wait until someone with a sense of using MNE Epochs and Annotations wants to actually do this sort of thing, and let that guide the API. Having an actual use case will make it clear what we need to add, I think.

[larsoner]

It would be nice to be able to set the annotations specifying epoch indices and times wrt to epoch times. But please correct me if I’m wrong.

Would you be able to illustrate an example for that use case?

A simple one is if someone does read_epochs_eeglab or creates an EpochsArray – it would be nice for them to be able to say “I know in epoch 2 in this time interval something bad happens”, both via interaction in the epochs browser, and programmatically via some suitable method like epochs.set_annotations_by_epoch(...) (name and API TBD)