Improve adjustment of search domains

[PhilippSchuette]

Currently we implement the most basic adjustment strategy imaginable for search (sub-)rectangles: If a root falls exactly on the rectangle boundary then the boundary gets adjusted. This should be improved as follows:

• Rectangle boundaries should already be adjusted if f'(z) / f(z) becomes too small (substitution in the argument principle integral shows that f'(z) / f(z) ~ 1 / d where d is the distance to the nearest root of f(z)).
• The strategy regarding how that particular boundary segment gets adjusted should be improved (the resulting rectangle should reliably be larger, not smaller).

[PhilippSchuette]

As part of this issue it might also be worthwhile (in terms of efficiency) to consider a more dynamic refinement strategy: Depending on the currently estimated total number of roots the refinement procedure could choose to produce not two but maybe four or even something like 2 * #number of roots rectangles.