Libhxl

Major changes:
- switching to using the Python requests library consistently, so that you can (e.g.) use the requests_cache package to get automatic disk-based caching of remote web resources
- added an ExplodeFilter for converting "wide" data to "long" data
- The Tagger can now take a default tag argument, to apply to every column without a specific tagging match.

- add a new Explode filter that changes series data to a more-normalised form (no command-line version yet)
- improvements to Add Column filter
- better number handling in Clean Data filter (can now handle exponential notation)
- add mask parameter to Merge Data filter
- start support for reading/writing filter chains encoded in JSON
- added add_attribute and remove_attribute methods for a dataset
- documentation and unit test improvements

Major changes in this release:
- use the unidecode library to ignore accented characters when comparing keys, etc. (should make merges much simpler)
- add a MANIFEST.in file -- should fix the bug where the default validation schema is unavailable after installing with pip.

Clean up for requirements and setup. Remove the dependency on Shapely, and make sure requirements.txt is in sync with setup.py. Also, many API doc improvements, and some minor code cleanup.

Removed the (stale and unmaintained) hxlbounds script, which allows eliminating the dependencies on libgeos and Shapely, simplifying installation. Will add a new repo for bounds support in the future.

Fix some bugs in I/O and the replace filter.