Autoreject - drop bad epochs

[eqian99]

• Solution: do local rejects + interpolation for epoch repair
  https://autoreject.github.io/stable/auto_examples/plot_auto_repair.html
  https://autoreject.github.io/stable/_modules/autoreject/autoreject.html#AutoReject.fit

• Why are we dropping fixed number of epochs (autoreject does this automatically)?

20% dropped is ok, but 767 epochs every signal dataset is sussy. Global rejection was not ideal.

• Additionally, look at repair https://autoreject.github.io/stable/auto_examples/plot_auto_repair.html#sphx-glr-auto-examples-plot-auto-repair-py

We are selecting epochs to remove with just a simple check that amplitude > 700. There are better ways to do this, namely autoreject:

https://autoreject.github.io/stable/index.html
https://github.com/facebookresearch/brainmagick/blob/main/bm%2Fautoreject.py
https://github.com/meeg-ml-benchmarks/brain-age-benchmark-paper/blob/main/compute_autoreject.py

• Task: implement that in fif-eeg-preprocessing.py