make write batch aotmic under txn sys

[letterbeezps]

Put data one by one makes WAL's atomicity ineffective. A batch of data modified by the same transaction should be regarded as a whole.

[skyzh]

While this looks like a valid fix, the original code is correct. If we think about how txn are used: the read_ts is always less than the txn to-be-committed, so no txn can actually read these new data. Note that we bump the commit ts at the end of the write.

I'll merge the fix and also add a new test-your-understanding question in the book.