* Correct behaviour during consumer rebalancing
* Correct behaviour for minor APIs like close(), pending(), seek() etc.
* Improved documentation

This can be made common for all producers later

The queues approach had a problem. Say the consumer is not
used to fetch messages for a while. In this while the allocation
has changed multiple times. When the next invocation of the consumer
happens, the oldest message is fetched from the queue. It will take
a while before the consumer catches up

The approach has been simplified with a shared memory protected by a
lock and notified using an event

Ensure that the standby consumer logic works
Also add a __repr__ method for Zookeeper consumer

Other changes
* If zookeeper consumer was run from the same process, kazoo Partitioner
  identifier would have been the same - resulting in errors. Fixed
* Add support for chroot to a Zookeeper path
* Remove dependency on simplejson
* Reduce the sleep time for the partitioner (in acquired state)

Conflicts:
	kafka/client.py
	kafka/conn.py
	test/test_integration.py