improvements in find_dividers function (#20)

[lomereiter]

Summary of changes:

• path links are represented as a numpy array throughout, of shape (number of links) x 2;
• each case in the original loop moved into its own function in utils.py; there are basic unit tests, too;
• each function operates on the whole array of path links and returns a boolean mask;
• in utils.path_dividers vectorized binary search does all the job (np.searchsorted)
• find_dividers now returns a pandas dataframe (from, to, path) instead of two nested sets;
• the sets are still reconstructed in segment_matrix, though (five lines of code)
• I discarded discard_useless_links function as it's dead code

[josiahseaman]

It's really encouraging to see progress here. I heard we're seeing a 30x speedup in load times. Just ping me when this is ready for review. I've got a whole dashboard of notifications, so I don't always notice right away. I wrote component_segmentation, so I know it well.

[lomereiter]

OK, this is ready for review. I'll create a separate issue for segment_matrix overhaul.