The availability restrictions that apply to user lists (group, grouping)
now apply in workshop:
* In user lists.
* When randomly allocating users (also now works as expected if you use
group mode and a grouping with the activity).
In a pretty rare case of zero variance of received assessments, the "distance"
of the evaluated assessment from the referential ("best") one was not increased
regardless it's actual value. This led to higher grades for assessments in
certain situations (see the tracker for a particular example).