When computing state for the dashboard, we fetch a list of all review

groups and then we fetch all starred review groups (excluding any IDs

we already fetched). On the user page, we fetch the user's list of

accesible review groups.
In each case, we unconditionally factor in Local Sites, whether or not

they're used on the server. We fetch all group state, even though we

don't need it all. We sort in the database. And on the dashboard, we

unconditionally check for starred review groups, even though we have

modern state for tracking whether they even have any.
This change optimizes the sidebar item computation to do the following:


Use LocalSite.objects.build_q() to only filter by Local Site if

   Local Sites are used on the server.


We now fetch the minimum amount of fields we need from the groups

   for each use.


We sort in Python rather than in the database, freeing up the

   database to give us the results in any order that's needed. This is

   going to be a small list, so it's cheap to do.


The dashboard sidebar now skips the starred review groups if we have

   computed state showing the user doesn't have any.


This results in simpler queries every time the dashboard or user page

loads, and reduces the queries for most dashboard users.
The existing dashboard unit tests have been updated to clear out any

starred review groups, to keep tests normalized and to avoid any

optional state in the computations. The exception is the sidebar test,

which does include joined groups.

Unit tests pass.
Tested viewing the dashboard and user pages.