Optimize summary writer

[zhangxin511]

Description

Older versions will try to create a logtail from the pending from Kafka first, then from db/alfred, lastly from the latest summaries. The second step doesn't consume very wisely because it added a lot of networkIO, and older logtail in latest summaries + the pending ones, should consist most of the tail already. We only need to make network IO calls when needed. So changed the logic to use the logtail from latest summary + pending from kafak, and determine any gaps, and then send networkIO request to fill these gaps.

Also, older version detects the gap is way too expensive, changes to use a simpler and easy version.

Breaking Changes

None

Reviewer Guidance