Broken boolean values

The way it is now, doing a cache.add has a slight race condition
between checking if the key exists already before setting.

This rewrite uses the native SETNX command to only set the key if one
doesn't exist yet.

Adding test to prove multiple connections are being formed when accessing multiple backends.  Thanks to @twidi for the report and jezdez for the kick in the pants.

Modified the timeout value semantics to better match the memcache cache backend, i.e. negative timeouts are instant expiration, 0 timeout is no expiration, and greater than 0 is time to live.

Bumped to version 0.9.0 -- Added ability to specifiy the parser class and the ability to use Unix domain sockets

Added the ability to use tcp sockets or unix domain sockets. Also added the
ability to specify a parser to use when processing responses from the redis
server.  To do this I had to bring back the custom connection pool, since
you need to instantiate a custom connection pool to be able to specify a
parser class.

Nudged to version 0.8.2.  Addresses handling of integers when using the 'get_many' method.  Fixes issue #20 on github.

Nudged to version 0.8.1, added ability to pickle the cache class. Added James Aylett to authors file.

Make cache object picklable.

To do this, we need to keep integers untouched while pickling any other type
of value. Existing tests should prove that the native incr is working, but
I added a test to show that if an integer was pickled and native incr is used,
then the cache backend will convert the pickled integer correctly.

The are two reasons:
* django already implements CacheClass instance as singleton, we don't
need to reimplement that
* connection pool in redis.py is already tested, we shouldn't have to
test it again
* default keywargs charset='utf-8' and errors='strict' are defaults in
redis.py, se we don't need to copy it over

Users should handle django.core.exceptions.ImproperlyConfigured better
then a custom exception with the same name