Comment 2 for bug 1190580

I am testing an alternative fix to the Mark's fix above:

1) Currently innobase_xa_prepare is returning early with success if fake changes enabled. I think it should return early with failure instead.
2) Then, if 2PC is in use, ha_commit_trans() will simply DTRT and not binlog the transaction.
3) But 2PC might be not in use due to the fake changes handlerton not being marked as RW, as noted by Mark's patch too. Thus, add a small loop before that iterates over transaction handlertons and bumps rw_ha_count for each fake changes handlerton.