The cbj would replay invalid records if the same write pattern (mkfs -> write -> mkfs -> write) repeats.
On the second mkfs(), current cbj wrongly recognizes the old records are valid beceause they have
correct seq. number at appropriate offset even though the records are expired.
To solve this, this commit introduces random value to identify the journal during mkfs()