cross_val_multiscore can crash with too few epochs

[larsoner]

I have a dataset with only 14 epochs and it crashes when calling cross_val_multiscore in decoding_time_by_time because there aren't enough epochs of each type:

(Pdb) w
...
> /opt/mne-bids-pipeline/mne_bids_pipeline/steps/sensor/_03_decoding_time_by_time.py(201)run_time_decoding()
-> scores = cross_val_multiscore(
  <decorator-gen-558>(10)cross_val_multiscore()
  /opt/mne-python/mne/decoding/base.py(880)cross_val_multiscore()
-> scores = parallel(
  /opt/mne-python/mne/decoding/base.py(881)<genexpr>()
-> p_func(
  <decorator-gen-559>(10)_fit_and_score()
  /opt/mne-python/mne/decoding/base.py(967)_fit_and_score()
-> test_score = _score(estimator, X_test, y_test, scorer)
  /opt/mne-python/mne/decoding/base.py(992)_score()
-> score = scorer(estimator, X_test, y_test)
  /opt/.../sklearn/metrics/_scorer.py(472)__call__()
-> return estimator.score(*args, **kwargs)
  /opt/mne-python/mne/decoding/search_light.py(333)score()
-> y = _fix_auc(scoring, y)
  /opt/mne-python/mne/decoding/search_light.py(771)_fix_auc()
-> raise ValueError(
(Pdb) p y
array([1., 1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.])

1. Should figure out why this step doesn't fail for the previous decoding step.
2. Should do something reasonable here -- probably set the scores to nan
3. Ensure whatever we do in (2) is handled properly by the average task