No need to account for revisions that are not current when calculating stokens
because those, by definition, are not the latest ones, and therefore won't have
the most recent stokens.
This becomes a problem when collections have many associated revisions.
The way were were doing it was implicitly sorting the query items and it
was causing millions of items to be sorted (even though the result should
only have one) making it slow.
By switching away from `get()` and `first()` we are telling django to
not try to sort.
