added handling of 3d numpy arrays and lists of uneven numpy arrays

[mitchellostrow]

The datasets I work with are typically of two possible shapes:

1. 3-d numpy arrays of shape (n_trials, n_samples, n_features) where n_trials denotes different initial conditions.
2. A list of 2- or 3-d numpy arrays of the above shape, but with different sample lengths.

We would like our DMD fitting procedure to be able to flexibly handle these arrays. The main functionality I've therefore added is the following:

1. detect the shape of the dataset, and generate x and y arrays [n_total_samples, n_features] accordingly. This can then be passed into the standard fitting procedure.
2. reshape the dataset back into their original shapes during prediction, so we can easily index into both.

I've added testing for each class as well to handle this separate cases. All tests pass and code has been properly formatted.