Discrete hypo test is too restrictive

[thehrh]

It currently forces the user to:

• specify some settings (at least a discrete selection, otherwise there's an exception raised since "there's nothing of substance to be done") for the 2nd hypothesis, even though one might just want to test one hypothesis against (fluctuated, Asimov) data
• run at least one "fiducial" fit, which makes no sense for the case of a single hypothesis tested (num_fid_trials is forced to be >= 1)

In brief, it seems too tailored to the NMO analysis (and thus probably scares off anyone not familiar with the NMO's intricacies), and is clumsy to work with in simple ensemble fits. It would seem I've been able to relax the above behaviour without introducing any issues into the analysis, so I'm planning to adopt the changes in my general_fitting branch.