refactor the boundary condition handling slightly

instead of passing a "minimal" fluid state that defines the
thermodynamic conditions on the domain boundary and the models
calculating everything they need based on this, it is now assumed that
all quantities needed by the code that computes the boundary fluxes
are defined. This simplifies the boundary flux computation code, it
allows to get rid of the `paramCache` argument for these methods and
to potentially speed things up because quantities do not get
re-calculated unconditionally.

on the flipside, this requires slightly more effort to define the
conditions at the boundary on the problem level and it makes it less
obvious which quantities are actually used. That said, one now has the
freedom to shoot oneself into the foot more easily when specifying
boundary conditions and also tools like valgrind or ASAN will normally
complain about undefined quantities if this happens.

examples/problems/fingerproblem.hh

@@ -177,6 +177,7 @@ class FingerProblem : public GET_PROP_TYPE(TypeTag, BaseProblem)
 
     enum {
         // number of phases
+        numPhases = FluidSystem::numPhases,
 
         // phase indices
         wettingPhaseIdx = FluidSystem::wettingPhaseIdx,
@@ -520,6 +521,14 @@ private:
         Scalar pn = 1e5;
         fs.setPressure(nonWettingPhaseIdx, pn);
         fs.setPressure(wettingPhaseIdx, pn);
+
+        typename FluidSystem::template ParameterCache<Scalar> paramCache;
+        paramCache.updateAll(fs);
+        for (unsigned phaseIdx = 0; phaseIdx < numPhases; ++ phaseIdx) {
+            fs.setDensity(phaseIdx, FluidSystem::density(fs, paramCache, phaseIdx));
+            fs.setViscosity(phaseIdx, FluidSystem::viscosity(fs, paramCache, phaseIdx));
+        }
+
     }
 
     DimMatrix K_;