Add feature importances support to iterated models

[OkonSamuel]

closes #66

julia> using MLJIteration, MLJBase, StatisticalMeasures

julia> X, y = make_moons(100, rng=123);

julia> iterated_model = IteratedModel(
           model=XGBoostClassifier(seed=123),
           resampling=Holdout(rng=123),
           measures=log_loss,
           iteration_parameter = :num_round,
           controls=[Step(2), Patience(2), NumberLimit(5)],
           retrain=true
       );

julia> mach = machine(iterated_model, X, y) |> fit!;
[ Info: Training machine(ProbabilisticIteratedModel(model = XGBoostClassifier(test = 1, …), …), …).
[ Info: final loss: 0.24434335559754639
[ Info: Stop triggered by NumberLimit(5) stopping criterion.
[ Info: Retraining on all provided data. To suppress, specify `retrain=false`.
[ Info: Total of 10 iterations.

julia> mach |> feature_importances
2-element Vector{Pair{Symbol, Float32}}:
 :x2 => 9.344354
 :x1 => 1.4283949

[ablaom]

Looks to me. Thanks for including the test.