Adds chosenFeatures for a more expressive DefaultStateRepresentation.

[gostreap]

The main objective of pull request is to provide a simple tool to choose which features to use.

chosen_features = Dict(
    "constraint_activity" => true, 
    "values_onehot" => true,
    "variable_initial_domain_size" => true,
    "variable_domain_size" => true,
    "variable_is_bound" => true,
    "nb_involved_constraint_propagation" => true,
    "nb_not_bounded_variable" => true
)

SR = SeaPearl.DefaultStateRepresentationTsptw{SeaPearl.DefaultFeaturization, SeaPearl.DefaultTrajectoryState}
learnedHeuristic = SeaPearl.LearnedHeuristic{SR, SeaPearl.TsptwReward, SeaPearl.FixedOutput}(agent; chosen_features = chosen_features)

We have also introduced a new statistic: numberOfTimesInvolvedInPropagation. This is a dictionary that associates to each constraint the number of times it is put in the fixPoint stack. This allows us to add the corresponding feature in the featurization of the variables.

[gostreap]

Because of a bad manipulation, this pull request has been duplicated, the review is in #215.