extend DESolver to use dithering

[mmckerns]

extend the DESolver to use dithering

From Rainer Storn:

I would like to draw your attention to the fact that
advanced versions of DE virtually always use dither in
their scaling factor to enhance diversity while keeping
the number of population members low. So if your
variable "scale" varies with equal probability
in the range [0.5, 1] rather than setting it to a fixed
value you usually get a more robust behaviour while
simultaneously increasing the convergence rate in the mean.
The reason for that is that in classic DE a reduction of the population
size increases the convergence speed but also the risk
to get stuck in a local optimum due to stagnation.
Stagnation again results from a lack of population diversity which
can be compensated by using dither. Usually dither is used
per generation - i.e. a changing value for "scale" in every
generation - and not per vector because it is simpler.
The results for both variants do not differ significantly.