The idea is to use the data obtained by "ANALYZE statement" (MDEV-406) for query optimization. In order for that to work, we will need to
1. Save the ANALYZE data somewhere where we could access it
2. Reuse that data in optimization

== Saving ANALYZE data ==

After MDEV-406, the data will be saved into the slow query log. The problems are

• There is no way to read that data back in the server
• Even if we implement reading, we will only be able to read the file sequentially.

== Reading and reusing the data ==

At the moment the idea is to re-use execution data for re-running the same query. Reuse will go as follows:

• We get a query
• We find ANALYZE data for this query (we could also try using ANALYZE data
  from the queries that matched the same query pattern)
• We optimize the query, making the optimizer to take into account the real
  table read costs and selectivities.
• This can result in the same query plan or in a different query plan..

TODO: elaborate further.

== Questions ==

• Do we need to save just the ANALYZE output (i.e. EXPLAIN + actual execution data), or it should be some other kind of data that can be reused by the optimizer? (current answer: just EXPLAIN)

• How exactly can saved data be used? (Current answer: we have precis selectivity/fanout numbers for the join order that was picked)

• Do we use data collection/re-feeding for all queries or for some of them only?