Mark Noble has made improvements to the work ID calculation algorithm (located in his code base at https://github.com/mdnoble73/aspen-discovery) to "correct some normalization issues and added a way to group records by providing alternate titles and authors". We should update our Python implementation to incorporate these improvements. While we're at it, we should also add test coverage of the various normalization algorithms, which include a lot of special cases that I don't think are documented anywhere.

https://jira.nypl.org/browse/SIMPLY-2775 is part of this work, but I filed it as a separate story because it's more urgent and much better-defined.