Revisit logic of UserScanQueryMatcher#mergeFilterResponse method

As HBASE-19729 comment , there are following points to be optimized:

1. when the number of cells exceed max version in scan, we should return SEEK_NEXT_COL match code, but if current code is INCLUDE_AND_SEEK_NEXT_ROW, we can optimize to choose the max step between SEEK_NEXT_COL and INCLUDE_AND_SEEK_NEXT_ROW, which is SEEK_NEXT_ROW
2. check to call ColumnTracker#doneWithColumn() only when we change the match code to SEEK_NEXT_COL or INCLUDE_AND_SEEK_NEXT_COL because of filterResponse.
3. More UT for those edge case..

https://issues.apache.org/jira/browse/HBASE-19729?focusedCommentId=16318131&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-16318131