Supplement functions in ReservoirTrajectory and BehaviorCloningPolicy

[peterchen96]

add update! and prob functions in ReservoirTrajectory and BehaviorCloningPolicy for solving problems mentioned in #386:

• use reservoir_trajetory to collect data for sl_agent
• replace average_learner with BehaviourCloningPolicy

So that I can define sl_agent as the following:

sl_agent = Agent(
    policy = BehaviorCloningPolicy(;
        approximator = NeuralNetworkApproximator(
            model = base_model |> _device,
            optimizer = Optimizer(sl_learning_rate),
        ),
        explorer = WeightedSoftmaxExplorer(),
        batch_size = batch_size,
        min_reservoir_history = min_buffer_size_to_learn,
        rng = rng,
    ),
    trajectory = ReservoirTrajectory(
        reservoir_buffer_capacity;
        :state => Vector{Int64},
        :action => Int
    ),
)