Building model indexes on up-to-date segments will delete the indexes built on the out-of-date segments

Currently, Kylin collects the out-of-date indexes from the up-to-date segments, rather than from all the effective segments, which will lead to the out-of-date indexes to be deleted. As a consequence, the query scanning on the historical segments will mismatch the indexes, to be pushed down to Spark engine directly.

The issue is simply to be solved, please see the attached pic.