A lot of stuff in NetworkUnitRay is general enough to be included in the base class NetworkUnitFeederInterface, so it doesn't have to be implemented again in deriving feature map feeders.
Currently NNUnit is synonym for class with ray feeder and activation function. However, feature maps may also want to use an activation function (e.g. in convolutional networks), but use different feeders. So we want an FedActivationUnit deriving from FedNetworkUnit and adding ACTF, and then NNUnit deriving from both FedActivationUnit.