Description
A user reported that Cassandra's estimate of memtable space used was off by a factor of between 3 and 10 for his counter columnfamilies.
We may or may not be able to fix the counter estimate (counter merging is a cranky best, and unlike normal merging can involve allocating new objects), but we can definitely use the SlabAllocator information to cap the error in our estimate at a fairly low amount.