Correct the handling of Dirichlet boundary conditions for the fully

implicit cell-centered models. The Dirichlet values are now evaluated on
the boundary and the resulting fluxes is added to the residual, instead
of evaluating the conditions at the cell centers and enforcing them in a
strong sense. 

The changes in detail: 
- CCElementVolumeVariables: in case of a Dirichlet boundary, calculate
  the VolumeVariables there and store them in the
  ElementVolumeVariables. 

- CCFVElementGeometry: the actual boundary conditions are not known
  here. Up to now, the information for the boundary faces has been set
  for outflow conditions. Since Dirichlet conditions should occur more
  often, the information is now set according to them. 

- CCLocalResidual: the function evalDirichlet_ is not needed anymore.
  All types of boundary conditions can be handled in
  evalBoundaryFluxes_. The handling of outflow conditions is now more
  complicated, since the FVElementGeometry has to be manipulated. 

- ImplicitLocalResidual: evalBoundary_ now differentiates between box
  and cell-centered. A dummy implementation of evalDirichlet_ is added. 

It works for the 1p2c and 2p tests. As expected, solutions change a bit.
Some of the problem statements can be simplified, since Dirichlet
conditions now are evaluated on the boundary for both the box and the
cell-centered models. All tests and reference solutions will be adapted
and committed separately.

Reviewed by Christoph. 



git-svn-id: svn://svn.iws.uni-stuttgart.de/DUMUX/dumux/trunk@10687 2fb0f335-1f38-0410-981e-8018bf24f1b0