Bug(fuse table): Insertion performance regression (seems proportional to the length of table history)

[dantengsky]

Summary

Observed that while keep ingesting data to the fuse table (at the same scale, single committer), the time of insertion keeps increasing, and seems to be proportional to the length of table history.

The above situation is NOT expected: the time of insertion (same volume of data), is supposed to be independent of the length of history.