Modify matching algorithm to reduce probability of rematching

[AndrewVSutherland]

It would be useful to incorporate an adjustment when computing the compatibility of groups during the matching process that subtracts points based on the number of group members who were previously members of the same group. One approach would be to subtract an exponential term c^r, where c is a a reasonably large constant and r in [0,1] is the proportion of former partners (when r > 0). It wouldn't take much to force a swap that avoids rematching when feasible.

This would reduce the probability that students who were previously matched and then left their groups would be placed in the same group, and facilitate classes in which students are explicitly instructed to periodically seek new partners.