* Redo setup.py, and remove ugly headers_workaround hack. Should result in fewer install problems.
* Update data downloading and installation functionality, by migrating to the Sputnik data-package manager. This will allow us to offer finer grained control of data installation in future.
* Fix bug when using custom entity types in Matcher. This should work by default when using the
  English.__call__ method of running the pipeline. If invoking Parser.__call__ directly to do NER,
  you should call the Parser.add_label() method to register your entity type.
* Fix head-finding rules in Span.
* Fix problem that caused doc.merge() to sometimes hang
* Fix problems in handling of whitespace