[cell-centered implicit] fix a bug in the handling of mixed boundary

conditions

Mixed boundary conditions here means that a part of the
equations/primary variables gets Dirichlet conditions, while the rest
gets Neumann conditions.

While in the cell-centered method pure Dirichlet conditions for all
equations are handled by calculating the resulting fluxes and adding
them to the cell residual, such a flux cannot be (easily?) calculated
for an equation that gets a Dirichlet condition as part of mixed
conditions. Therefore, such a Dirichlet condition is implemented in a
strong way by _replacing_ the corresponding cell residual.

It is important that this replacement is done at the very end of the
residual calculation. However, for corner cells this has not been
guaranteed so far. Therefore, fluxed resulting from the other boundary
parts of a corner cell could have been added to the replaced residual,
obviously leading to a wrong boundary condition treatment.

This patch resolves the issue by guaranteeing that the residual
replacement is done at the end of the residual calculation.

Brought to attention and reviewed by Thomas. Thanks.



git-svn-id: svn://svn.iws.uni-stuttgart.de/DUMUX/dumux/trunk@13823 2fb0f335-1f38-0410-981e-8018bf24f1b0