Include a new two-sided test and associated FDR estimation

[magnusrattray]

The current DE method in BitSeq, pplr, is a one-sided Bayesian test (testing FC<1 or FC>1) and it is therefore difficult to associate gene lists with an FDR as is standard with other DE tools. We could instead include a two-sided test by making the null hypothesis a FC range (e.g. 1/2>FC<2) within which changes are not considered large enough to be of interest.