We previously processed batch jobs round robin, i.e., one item
for each job per round. This is fair from the job point of view,
but not from the user point of view when one user has many jobs.

We now process batch jobs one item for each user per round,
where we pick the oldest job if a user has more than one. Users
are defined by their email address.

Batch jobs submitted via the webservices all have the same email
address, so they are effectively throttled as if all from the
same user. Adapting the webservices to also allow setting an
email address is future work.