handle edge condition on recoonect - repeated network errors while pa…

[gleonid]

this is a continuation of #420
I have realized that there is a scenario when transaction will get skipped

Consider following sequence of events:

1. we have processed transaction A (TranA) and encountered a network error while reading events for next transaction (TranB). At this point b.prevGset equals TranA and b.currGset equals TranB
2. we get to BinlogSyncer.retrySync and re establish binlong stream "from TranA". Thus first transaction we are going to see - TranB (which we already saw before error was encountered)

Thus we are going to enter advanceCurrentGtidSet with state:
b.prevGset === TraA
b.currGset === TranB
gtid === TranB

without proposed change - upon exit of this function we will have state:
b.prevGset === TranB
b.currGset === TranB

Which means, that if network error is encountered again, while processing same TranB - we will be re establishing replication stream from TranB, in other words first transaction delivered after such reconnect will be TranC, so TranB will never be fully processed.