Kernel*Distances: Fix multiple keys

[yannikschaelte]

Currently, some use the mean argument wrongly, and set the dimension incorrectly. E.g. for IndependentNormalKernel, the variances should be initialized for each key separately, e.g. something like

var = {'s0': np.array([1,1,1]), 's1': np.array([1,1])}

[yannikschaelte]

otherwise we have wrong normalizations. this did not affect the anylsis in the past, but nonetheless is wrong and easy to misuse.

[yannikschaelte]

done in #197