Extend Storage Engine API to include estimate of full-text search

When the optimizer looks for a best execution plan, it needs to know how many rows a particular condition can match. For historical reasons a full-text search query that uses a full-text index is always assumed to return 1 row.

While it is kind of correct for a NL full-text search, it is definitely wrong for a boolean full-text search.

For storage engines like Mroonga to be able to provide the optimizer with an estimate of the row count of a full-text search the Storage Engine API need to be updated with functions for the storage engines to provide the estimate.