Reiterate runs for uncertainty intervals

[SdgJlbl]

Building up on feature #48, if the experiment contains some stochasticity (random train / validation split, random initalization, stochastic gradient descent), we want to reiterate several runs of the experiments and to save the resulting metrics.
Theses metrics values can be exported as a raw csv file or in a aggregate way (to estimate uncertainty).