fix consumer retry logic (fixes #135)

Fixes bug in the follow condition:

* Starting buffer size is 1024, max buffer size is 2048, both set on an instance level
* Fetch from p0, p1 and received response
* p0 has more than 1024 bytes, consumer doubles buffer size to 2048 and marks p0 for retry
* p1 has more than 1024 bytes, consumer tries to double buffer size, but sees that it's at
  the max and raises ConsumerFetchSizeTooSmall

The fix changes the logic to the following:

* Starting buffer size is 1024 set on a per-partition level, max buffer size is 2048 set on an instance level
* Fetch from p0, p1 and received response
* p0 has more than 1024 bytes, consumer doubles buffer size to 2048 for p0 and marks p0 for retry
* p1 has more than 1024 bytes, consumer double buffer size to 2048 for p1 and marks p1 for retry
* Consumer sees that there's partitions to retry, repeats parsing loop
* p0 sent all the bytes this time, consumer yields these messages
* p1 sent all the bytes this time, consumer yields these messages

0 files changed, 0 insertions, 0 deletions