seems like there are "stable" choices of delta: as soon as delta is
small enough, iteration numbers stay within a narrow range on every
level. else, they seem to explode at some point

Number of Kacanov steps seems to blow up

* exact prolongation operators for general finite element functions are
  available now
* major clean-up of mesh update during refinement, introducing new hooks
* minor bugfixes in plotting and RGB refinement

weights were computed in a wrong way -> now: same way as in
nodalInterpolation

Bug occured if 'SubdivisionFactor' is given and in the subsequent loop
only one element is drawn. Then, the broadcasting sub-routine returned
an array with wrong dimensions.

Testing was done by switching specialized projections in goafem /
iterative linearization example to GeneralFeProjection

Sorting the edges in RGB according to length only affected edge marking,
not the actual refinement step. This led to wrong refinements if the
longest edge was not the refiment edge.

Reordering edges is now outsourced to a method of mesh to do it in a
global and sustainable way. This also cleans up RGB a little.

* maximum criterion (eta(T) >= (1-theta)*max(eta))
* equidistribution criterion (eta(T) >= (1-theta)*mean(eta))