Investigate how to handle the deleted document in fast filter optimization

[bowenlan-amzn]

Follow up task for #9310

The filter rewrite optimization in date histogram doesn't account for the deleted documents. We would fall back to default aggregation execution path if deleted document exists. However, there are situations where deleted documents are hard to be merged away. e.g. the segment is large (~5gb) but the deleted documents counts are relatively small. So our optimization will never be able to kick in under such scenario.

[bowenlan-amzn]

As discussed in an internal review meeting, it’s probably better to look into the other way — how to clean up deleted documents of big segments easily, instead of hanlding in our optimization path.

Currently this is not in plan.