Reviews and review request queries attempted to pre-fetch a list of IDs

of review groups and repositories they had access to, in order to

optimize the main queries. This didn't work quite right, as the return

type provided by Django wasn't a straight list but rather a type of

queryset, which it was turning into a JOIN when passed into the main

queries. That created a pretty deeply-nested query, which was far from

optimal.
This wasn't helped by one of subqueries for review requests calling the

wrong function, pulling in the actual subquery and not the ID list.
The queries now use the functions for ID lists in all cases, and those

explicitly return IDs to avoid the issue.
The result is that the ID queries are separated out, turning this one

giant query into 3. The ID queries are small enough and simple enough to

be cached on the database end, speeding those up.
There's still room for improvement here, which is being explored. Some

caching and pre-fetching of state like the existence of Local Sites,

private repositories, or invite-only groups will let us further simplify

the queries for all users.

Unit tests pass.