Add averaging 1D learner

Add a learner that can learn a function that returns a random variable.

This is like the 1D learner problem, but with the additional possibility of re-sampling the function at an already-seen point, x, in order to get a better estimate for the expectation value of f(x).