arg_max optimisation

[boristim66]

Hi,
thank you for the wonderful program,
I sucsessfully use its algorithms rewritten in C++ in my work.
I would like to draw your attention to the fact that the argmax_clusters function works very inefficiently with methods "xxx_clade". Each such method, at first, calculates the required distances when traversing the tree without using the "threshold" parameter, and then the another algorithm (the same in all "xxx_clade" methods) performs clustering with this parameter.
If you separate the distance calculation (one times) and clustering in cycle with "threshold" stepping of these methods in the argmax_clusters, you will get at least double/triple acceleration of this function, and when using the min_clusters_threshold_med_clade method, this acceleration is even difficult to estimate.
If you find time for such optimization, users will be very grateful to you.