Regression in 3.0, breaking the fix from CASSANDRA-6069

The goal of the fix of CASSANDRA-6069 was to make sure that collection tombstones in an update in CAS were using t-1 because at least in INSERT collection tombstones are used to delete data prior to the update but shouldn't delete the newly inserted data itself. Because in 2.x the collection tombstones are normal range tombstones and thus part of the DeletionInfo, we went with the easy solution of using t-1 for all of DeletionInfo.

When moving that code to 3.0, this was migrated too literally however and only the DeletionInfo got the t -1. But in 3.0, range tombstones are not part of DeletionInfo anymore, and so this is broken.

Thanks to aweisberg for noticing this.