Broker startup could be inefficient when using old message format

During broker startup, we call `Log#recoverSegment` when we find corrupted indexes, for segments beyond the last check-pointed recovery point, and for any ".swap" segments created by log cleaner. One of the things `Log#recoverSegments` does is to build up the producer state, starting from any previous snapshot file that is available. For logs using message formats older than V2, we could skip building up this producer state which would essentially speed up recovery.