Speed up the computation of the partitions matrix when no haploweb is required

[jflot]

Right now this computation is very slow, which is a big problem when dealing with thousands of markers (e.g. RADseq).