changed IDifferentiable interface to include getter providing neurons with activation function based on derivatives of the parental activation function
added neurons representing 1st and 2nd derivatives of the logistic neuron

-added a new test showcasing the use of subnetworks sharing edge weights and multiple error functions in swarm optimization

-added an example testing multi-network error function

FIX: Fixed missing initialization of 'n_activation_function_parameters' variable in the constructor.

ENH: Moved derivative methods to IDifferentiable class-interface, so that not all activation classes have to implement them now.