Jug

* Fix use of numpy subclasses
* Fix redis URL parsing
* Fix ``shell`` for newer versions of IPython
* Correctly fall back on non-sqlite ``status``
* Allow user to call set_jugdir() inside jugfile

* Faster decoding
* Add jug-execute script
* Add describe() function
* Add write_task_out() function
* Allow for set_jugdir() to be called inside jugfile

* Added debug mode
* Even better map.reduce.map using blocked access
* Python 3 support
* Make sqlite3 a soft dependency for ``status --cache``
* Load config when embedding ipython
* Documentation improvements

* Add CustomHash wrapper to set __jug_hash__
* Print traceback on import error
* Exit when no progress is made even with barrier
* Use Tasklets for better jug.mapreduce.map
* Use Ipython debugger if available (patch by Alex Ford)
* Faster --aggressive-unload
* Add currymap() function

* Fix parsing of ports on redis URL (patch by Alcides Viamontes)
* Make hashing robust to different orders when using randomized hashing
(patch by Alcides Viamontes)
* Allow regex in invalidate command (patch by Alcides Viamontes)
* Add --cache --clear suboption to status
* Allow builtin functions for tasks
* Fix status --cache (a general bug which seems to be triggered mainly by
bvalue() usage).
* Fix CompoundTask (broken by earlier __jug_hash__ hook introduction)
* Make Tasklets more flexible by allowing slicing with Tasks (previously,
slicing with tasks was **not** allowed)

* More flexible mapreduce()/map() functions
* Make TaskGenerator pickle()able and hash()able
* Add invalidate() method to Task
* Add --keep-going option to execute
* Better help messsage