Related to #42

Adds new ConsumerFetchSizeTooSmall exception that is thrown when
`_decode_message_set_iter` gets a BufferUnderflowError but has not yet
yielded a message

In this event, SimpleConsumer will increase the fetch size by 1.5 and
continue the fetching loop while _not_ increasing the offset (basically
just retries the request with a larger fetch size)

Once the consumer fetch size has been increased, it will remain
increased while SimpleConsumer fetches from that partition

Was hard coded to 1024 bytes which meant that larger messages were
unconsumable since they would always get split causing the consumer to
stop.

It would probably be best to automatically retry truncated messages with
a larger request size so you don't have to know your max message size
ahead of time

Conflicts:
	kafka/__init__.py
	kafka/consumer.py
	test/test_integration.py

This alleviates IPv4 -vs- IPv6 issues in ZK and Kafka.