Optimize AffinityPropagation class

[Markoy8]

Added a class extending the existing AffinityPropagation.java class.

Its implementation upgrades the previous one in terms of computation efficiency. The two core functions are reimplemented. Computational complexity reduced from n^3 to n^2 for both the functions.

Moreover it adds the computation of the median of similarities if the "selfDivergence" parameter is not specified in input by the client.

The new implementation could substitute the existing one. It obtains exactly the same results, in a much shorter computation time. (With a local test dataset it is 28 times faster)