pull out a subset of the dataset by reducing the number of columns in the last dimension. This does not reduce rank. This assumes the dataset has no row with length>end.