* according to Rene

* this defines CHECK_CRTP and CHECK_AND_CALL_CRTP
  (has still to be made threadsafe perhaps?)
* defines CRTPInterface

* there is now DUNE_THROW_COLORFULLY :)

* use one template which either redirects scalars to
  dune-common and checks vectors componentwise
* this template is then generated for eq, ne, gt, lt, ge and le

* LA::VectorInterface
* LA::EigenDenseVector

* rename DSC_TOSTR -> DSC_FRSTR
* group toString/fromString

sum and maximum value need to be computed on _every_
process and may then be output only on rank 0.

using c++11 string conversion for special cases avoids icpc 14 bugs

* fixes #252

* Now this one is a little tricky. The problem is that we
  basically need a specialization for every combination
  of dimensions (since some functions may only be
  implemented for some dimensions).
* We now generate specialized declarations for all
  dimensions in the header.
* We also generate specialized definitions for all
  dimensions in the source and we require the
  implementer of a new functions to add the function in
  the source file.
* Adding a new function is a sub optimal task, since it
  requires some copying. This solution may not be the
  most elegant but it seemed to be *fastest*, since we
  can just copy/paste find/replace.
* Ideas and patches are welcome!