add versions of experiment configurations

[thartbm]

By having versions of experiments that only change 1 or 2 parameters here and there (the number of training trials in 1 subtask, the rotation size somewhere, or the speed at which the rotation increases), we can more easily (with fewer human error) keep everything else exactly the same.

Versions can be picked at random or they can be counterbalanced based on how many have already been done in each version.

The latter would require the experiment runner to be aware of the data to some extent.