Inconsistent use of discount factors #39
Description
We're currently not consistently using discount factors.
Our implementations of ApprIRL and RelEnt IRL do use a discount factor. For example, ApprIRL works with discounted feature counts. By contrast, MaxEnt IRL doesn't (the MaxEnt class is initialized with a discount factor, which is set as an attribute, but the discount factor is never used in any computation).
We inherited these inconsistencies from the original papers. E.g., the MaxEnt paper doesn't mention discount factors and implicitly assumes an undiscounted setting.