[pyclustering.cluster.kmeans] Optimization via numpy

[annoviko]

Introduction
Current implementation does not use numpy. If CCORE is not used then numpy should be used for that purpose to increase performance of the K-Means algorithm.

Description

• In case of Python numpy should be used for calculation.

[annoviko]

The first version of K-Means using numpy. Results:
With numpy optimization:

Execution time (1000 2D-points): 1.1295836647650328
Execution time (2000 2D-points): 3.0258542384768683
Execution time (3000 2D-points): 3.934701825511546
Execution time (4000 2D-points): 13.279997776200211
Execution time (5000 2D-points): 14.245599869993242
Execution time (10000 2D-points): 16.167380278317097
Execution time (20000 2D-points): 72.8136846366796

Without numpy optimization:

Execution time (1000 2D-points): 0.0009254428355157932
Execution time (2000 2D-points): 0.001377045254325718
Execution time (3000 2D-points): 0.0019560885072316255
Execution time (4000 2D-points): 0.0025690589620557033
Execution time (5000 2D-points): 0.00314924262511012
Execution time (10000 2D-points): 0.006151372341062462
Execution time (20000 2D-points): 0.012956769902295356

numpy based implementation is faster than current. Optimization is accepted.