Make get_shifts return value more useful; add test for shifts

[ethanbb]

Fixes #310. This is an API change for mcorr.get_shifts(). Changes in this PR:

• Don't return the first output of get_shifts anymore, since it was confusing/seemingly not very useful
• For piecewise case, return an n_dims-length list of n_frames x n_patches matrices rather than an n_dims*n_patches-length list of n_frames-length vectors.
• Remove the default value for pw_rigid, since it seemed too easy to forget to pass this and get an incorrect output as a result (came up when I was running the new test)
• Test the shifts against ground truth (currently only tests the piecewise case). I made a new zenodo upload with a shifts ground truth file. Note that this is based off of the upload for Don't filter out NaNs for contours #309, so since that's not merged yet, the contour ground-truth tests will fail.