Move random choice logic to the backend and test

[goldpbear]

Currently, both the random selection of a respondent and the random selection of the question presented to the user on the question2 stage of the experiment are determined client-side, in the JS code.

Instead, I think we should make the random determinations in the backend. That way, we can write tests that check we are getting the expected statistical distributions of random questions over many runs of the experiment.