Optimize DISTINCT and GROUP BY for key prefixes

If a table `test` has an index on a column `c0`, then the following queries can all be optimized by an engine that is aware of the tree structure of tables:

- `SELECT DISTINCT c0 FROM test;`
- `SELECT c0, COUNT(*) FROM test GROUP BY c0;`

These are both queries that a user might reasonably want to know about an indexed column: what different values does it have, and how many rows have each of these values? These are also both things that can be efficiently computed with a tree walk: the algorithm can detect when the key column doesn't change inside a block and read that block's count instead of recursing into it.

However, we don't currently optimize either of these queries.