Add a test for stability of the value function (1)

[wasowski]

If two value functions are close they yield the same policy. This can be done by just checking whether in each state there is at least a certain guaranteed distance between the first and the second best action.