Describe the bug
In elastic.py, the vanilla python implementation of DTW apparently doesn't handle variable length time series properly. I believe it assumes the cost matrix is square, which it will not be for time series of different lengths. https://github.com/alan-turing-institute/sktime/blob/74c14cbdf01f5acb55294f69738e03facf09c80f/sktime/distances/elastic.py#L54

Expected behavior
The cost matrix is sized corresponding to each of the input time series allowing for variable length time series.