Range-slice queries with CL>ONE do unnecessary read-repairs

Scenario:

• keyspace with RF>1
• table filled with data (no tombstones)
• CL QUORUM (anything > ONE)
• select * from table (range-slice)
• trace output looks fine
• create some tombstones (e.g. delete from table where key = foo)
• select * from table (range-slice)
• trace output shows mutations (RR)

That RR is triggered from RangeSliceResponseResolver.Reducer.getReduced.

Seems like a live-cell-count comparison is missing.

Will create a dtest to prove the behavior and probably a patch.

NB: RowDataResolver and RowDigestResolver are fine.