Audit all strings in djblets and convert to __future__.unicode_literals
This change represents an audit of all of the strings in the djblets codebase.

Where the strings were actually binary data (specifically with our treatment of

settings.EMAIL_HOST_USER and settings.EMAIL_HOST_PASSWORD, and anything passed

into sha1() or md5()), I've made it use the bytes type. Otherwise, these are

now unicode objects by virtue of importing unicode_literals from __future__.
There's also a bunch of fixes to our handling of cache keys, especially those

for large data, much more consistent. We always keep the cache key as a unicode

object and create derivative keys (like the chunk keys for large data) with

that. At the last moment, before calling cache.get() or cache.set(), we pass

that to make_cache_key, which will shorten it if necessary using md5, and then

encode to utf-8 (since not all cache backends support the unicode object as

keys).

Verified my assumptions about treatment of unicode vs. str in getattr,

hasattr, delattr, __getattr__, __hasattr__, __delattr__, hash, regular

  expressions, and memcache keys.
Ran djblets unit tests
Ran Review Board unit tests