Assign test configurations in a more intelligent way, ensuring balance

[G-Thor]

The current method of assigning configurations to listeners in Latin Square MOS tests uses the number of people who have started the test as its basis. This avoids the race condition when two people are performing the test at the same time but introduces "gaps" when someone starts a test and doesn't complete it.

It would be better to include an additional check whether some configuration has fewer completions than others and prefer those configurations.