[Wal] The deconstructor of FileBaseWal class could cause an error in some cases about nebula

[dutor]

I can confirm this bug.
To conclude what @zlcook described:
Both of the exceptions are due to that the rollback logic didn't take care of the case when the target log id resides in the latest incomplete buffer. To resolve this problem, upon rolling back, we need to send the lastest buffer to flusher, if and only if the buffer has not been flushed out. Of course, the buffer also needs to be truncated before persisted.

I will leave this issue to @sherman-the-tank to fix.