If number of epic ids we pass to BulkEpicAggregateLoader reaches a limit
(~200), then postgres uses a different plan for the query which is much
slower.

To avoid this situation, we get this metadata in two phases:
1. get epic ids for epics and its descendants (we use pluck because
this query is recursive and find_each wouldn't help here)
2. we iterate in batches and get metadata for each of batch