Improve DISTINCT queries on PostgreSQL via recursive CTE

[bubiche]

Attempt to improve performance issues with PostgreSQL as stated here. Follow up of #397

PostgreSQL lacks MySQL's Loose Index Scan, so DISTINCT queries on queue_name degrade to full index scans on large tables. The recursive CTE walks the B-tree index jumping between distinct values.

I have encapsulate the database specific logic in Record::LooseDistinct so the rest can just use distinct_values_of, which uses recursive CTE to emulate loose index scan on PostgreSQL and normal DISTINCT query for MySQL/SQLite.

This should help with:

• Worker polling with prefixes/wildcards (in QueueSelector). We have had situations where solid_queue_ready_executions exploded even though it's normally small due to issues in other parts of the system and this would have helped.
• Mission Control (in Queue.all), this should solve Slow DISTINCT query for finding queue names with Solid Queue mission_control-jobs#160