Unify "activation" and "derivative" arrays-of-pointers-to-functions, into one

[GiorgosXou]

I had this great idea that probably will reduce sketch-sizes who utilize multiple-activation-functions-per-layer-to-layer when backpropagation is enabled, by just merging

static const method_function derivative_Function_ptrs[NUM_OF_USED_ACTIVATION_FUNCTIONS];
static const method_function activation_Function_ptrs[NUM_OF_USED_ACTIVATION_FUNCTIONS];

into one (using just a constant offset).

https://github.com/GiorgosXou/NeuralNetworks/blob/master/src/NeuralNetwork.h#L939-L947

[GiorgosXou]

This won't effect the performance and the overall resources of NNs with disabled backpropagation