Personal view about 3.6 Partial SSI

[Beauuu-0826]

In my view, task 2 https://skyzh.github.io/mini-lsm/week3-06-serializable.html#task-2-track-read-set-in-scan here isn't able to implement(it might be wrong), cause I can't iterate all keys given a RangeBound, thus the testcase of task2 can't be cover also.
And I've noticed that the bonus task mention about predicate lock, and this might need to change the data structure used to track read set and write set to avoid the write skew phantom, like track read-op as share mode, write-op as exclusive mode and do some conflict detections

And when I implement 3.6, I think it's better we can move the transaction layer up into MiniLsm instead of LsmStorageInner, then we can keep the LsmStorageInner implementation unchanged, and since we gain the commit lock to enforce only one transaction can enter the commit phase at a time, the commit_ts can simply be mvcc.latest_commit_ts().

[skyzh]

thanks for the feedbacks :) noted and will fix them in the next revision