now importance values have same shape as tensors without the batch dim

used to define tensors for ML training
currently temporary solution to allow for tuple-like use as it was before and class-like use as it is now

[recipes] common: fixes blackout (higher order slice is an iterable itself so we can not make this optional)

if somebody needs non-default behaviour, we should do this in another way!

used to split dss object in train valid and test according to event number
first map prepare:
  1. duplicates data ([012]->[01201])
  2. saves correct chunk starts and stops:
    split 0: test=[0], valid=[1], train=[2]
    split 1: test=[1], valid=[2], train=[0]
    split 2: test=[2], valid=[0], train=[1]
then map apply