KS with mx.diff=TRUE (GSVA-like)

[tetomonti]

It would be worthwhile to implement a version of the score similar to the one implemented in GSVA, where the difference between the top positive peak and the top negative peak is computed. Thus, if a signature is half enriched at one end and similarly half enriched at the other end, we will get a score close to zero, rather than the largest of the two, which would be misleading.