Bloom filters and column families

[mdcallag]

I assume we will put many indexes in one column family as column family per index doesn't scale to a large number of indexes. When that is done we need a method for specifying the index columns that will be used in the column family. This assumes that we can use bloom filters on a prefix of the index key and that the format of an index key within RockDB is:
<index ID - hidden column> <index column 1> <index column 2> ...

In that case the options are:

1. no bloom filter
2. bloom filter on index ID
3. bloom filter on index ID, index column 1, ..., index column N-1>
4. bloom filter on index ID, index column 1, ..., index column N>

Note that

• 2. and 3) above are on a prefix of the visible index columns
• 2), 3), and 4) require all indexes in the column family to have a common prefix in columns

This requires much more discussion.