Refactoring of the prediction of Gaussian filters.

It moves the performPrediction() from SampleBasedGaussianFilter to GaussianFilter. This also
includes setUseAnalyticSystemModel() and getUseAnalyticSystemModel(). Subclasses of
GaussianFilter now have to implement the predictedMoments*() methods. The
SampleBasedGaussianFilter class is removed and the class
SampleBasedJointlyGaussianPrediction is introduced to implement the prediction using samples
and assuming a jointly Gaussian distribution of state and noise (like it is done, e.g., by
the LRKFs). LRKF and PGF now inherit from this class, while GPF inherits from
GaussianFilter. All affected filters and interfaces are updated according to all these
changes.

In doing so, this allows for a much better understanding of the code, code reuse is
improved, and also the EKF (that did not inherit from SampleBasedGaussianFilter) can now use
analytic moment computation for the prediction.