fix rows vector according to pyimagesearch python source code

[arashrahmani]

implementation of rows vector was not like the PyimageSearch article (it was incomplete) and if you test the previous version of code on multiple objects you can see that it has some bugs. the python code is as follows: rows = D.min(axis=1).argsort()

the rows vector is calculated in sorted order of arguments based on the values of the D matrix. so I used sortbyfirst function (in C++) instead of numpy.min to get the minimum distance in axis1 and a for loop that gives the second attribute of each pair(arguments) of the sorted D instead of argsort.