[FIX] Rewritten serialization of ExprtkWrapper, so that Boost.serialization doesn't have to be friended anymore

[ENH] NeuralNetworkSum class now implemented using PIMPL to remove Boost::serialization header dependency induced by ExprtkWrapper.

ADD+MOD+FIX: added new evaluation functions for debug/testing purposes, modified random generation of parameters, fixed the calculation of gradient via backpropagation

ADD+MOD+FIX: added new evaluation functions for debug/testing purposes, modified random generation of parameters, fixed the calculation of gradient via backpropagation

ENH: Only classes belonging to API are now contained in 4neuro.h. All those classes were enclosed in 'lib4neuro' namespace.

ADD: added preliminary version of gradient calculation vie backpropagation, added supporting methods to error classes and learning algorithms utilizing gradient calculation

ENH: Removed both linking and include dependencies from examples => lib4neuro can be used just by linking it and including its header without any external dependencies needed.

FIX: fixed an issue with ExprtkWrapper being incorrectly destryed before being used in the NetworkSum class

MOD: Modified examples and unit tests to account for the changes