Rprop improvements

[abrock]

I added some functionality to the maximize() function in RProp:

• Print the current set of hyperparameters in every iteration if the user wants that. Default behaviour is unchanged.

I checked the behaviour of the function, especially what happens if I allow a large number of iterations (200). I found that after the first 50-100 iterations the method gets stuck and no updates to the best_params vector occur.
I therefore implemented a cooldown strategy:

Merging this would make #9 and #10 obsolete.

• If the new value of the target function is not better than the old decrease stepsize_factor which is multiplied with the calculated gp->log_likelihood_gradient. Initial value for stepsize_factor is 1.0 so for the first iterations the behaviour is unchanged.
• If the stepsize is smaller than a threshold (default: 1e-2) stop