Implement confusion matrix in mulan

[ghost]

P1: One point is that if you implement the matrix using separated labels its dimension is L+1xL+1, where L is the number of labels, then the sum of confusion matrix will be greater than the number of instances because of the multilabel classification, in contrast to weka where it is equal to the number of instances. 

P2: Other point is that the confusion matrix can be implemented using the number of distinct labelsets. And thus the sum of matrix will be equal to the number of instances, but the matrix will be much bigger than the above.