known bugs:

* evaluateFactor() in integralterm.hh is called with the *local* position
  of a given quadrature point, this should be first transformed into
  global coordinates.
* PQkTransportBasis cannot be used as a test space at the moment.
  There are compile-time errors that are probably caused by using
  ChooseQuadrature without specifying the transport direction β.
* Bug in convergence_test.cc with refined a posteriori FE:
    * a posteriori error is too large by addition of a constant under
      the square root. This constant is
      * 3*(f^2) for 1 refinement.
      * 15*(f^2) for 2 refinements.
      Assumption: The summand for level l and rhs f is
                  (N(l)-1)*(f^2),
                  where N(l)=4^l are the number of subcells
    * Check scaling of gradients in refined FE!
    * Compare system matrix (with OptTestSpace) of dg-refined test search space
      of different refinement levels and the old one using subsampled FE.

technical debt that needs to be paid off sooner or later:

* Replace boost::fusion with dune/common/tupleutility.hh or use
  dune-typetree instead of std::tuple.
* Beginning with version 1.58 boost::fusion has better support for std::tuple,
  thus we can maybe get rid of those boost::fusion::vectors.
