Inactive partitions break fast-forwarding

[kbourgoin]

This is a fairly specific case, but one that should be handled:

• Partitions are set to roll over based on time, not size
• Nothing is written to a partition for longer than the expiry (all logs have expired)
• A samsa client tries to connect and it's latest offset is no longer in kakfa (requiring fast-forwarding)

In this case, the client fast-forwards, but has nothing to read and ends up entering an infinite loop where it's constantly fast-forwarding but not getting anywhere.

I didn't investigate long enough to determine if it's an actual loop or it's just that every time next_message is call it prints the log warning. It may just be that the message needs to be suppressed in this case.

[kbourgoin]

I've reworked some of the code around fast-forwarding and generally handling OffsetOutOfRange errors. The issue's been taken care of. Verified by testing the case manually.