offer randomized contrast

[christofs]

Offer the option to run a comparison not on a meaningful partitioning of the data, but on a random one, for a better understanding of the level of differences to be expected if the partitioning is not meaningful.

[christofs]

Partly implemented.

Based on the current implementation, it would be very interesting to do this multiple times internally and calculate list of zeta score distributions ranked by mean or median, based on such multiple random partitionings of the data, then use this to do significance tests on the zeta scores with meaningfully partitioned data to estimate which zeta scores can actually be considered statistically significant given a certain text collection and partitioning.