These two files are very similar, and merging them will reduce
maintenance effort in the future.  As of this patch, there is
now quite a number of preprocessor directives in the new merged
cosserat-continuum.cc.  That is not pretty, but the new state
is still much easier to maintain than two completely separate
files.  What is more, I hope to be able to reduce some of that
preprocessor trickery by smarter use of the dune-functions
basis trees.

This used to work only with dune-fufem bases, but the method
constructBoundaryDofs recently learned to also handle the bases
of dune-functions.

This used to be a yes/no, parameter, now we require 'cube'
rather that 'yes' or 'true'.

The code it contains is adolcnamespaceinjections.hh.

Capturing any polymorphic object by value is wrong,
and gcc 8 started to warn about this in the case
of exceptions.

The latter is deprecated as of dune-functions 2.6.

This is needed to build with a recent dune-common and dune-fufem

gcc warns about some of those nowadays.

Otherwise some dune-common stuff will not work properly,
for example copying of FieldMatrices that use adouble
as their number type.

The method indexSet has been removed from dune-functions
a long time ago.

This is planned to be used to start a fine-grid computation on the coarse-grid
solution, or to start a high-order computation from a low-order one.  This is
helpful, in particular, for measurements of the discretization error, which
take a lot of time.

Previously, the NonplanarCosseratShellEnergy had the normal field of the unit sphere
hard-wired in the code.  With this patch, using general triangulated surface as
reference geometries is possible.

This allows to select the energy at run-time, but this is not what we do.
It also allows to give extra parameters to the nonplanar Cosserat energy,
and that is what we need: in a subsequent patch the main program will compute
a surface normal field and pass it to the Nonplanar Cosserat energy.

We can now completely control the approximation order by a const int,
which is much nicer.

This data is used by the compute-disc-error program to measure discretization errors.

... with values expected to be given in the parameter file.

Scaling the trust region is the key to success, because in general the midsurface
deformation and the rotations will have values in different ranges.

That really was a hack: the build system should do it.

[[Imported from SVN: r10111]]

[[Imported from SVN: r10103]]

[[Imported from SVN: r10098]]

Seems to be a bug in IPOpt, or in our build system

[[Imported from SVN: r10086]]

This is much cleaner than handing over the Grid only, and then trying
to guess the basis using the preprocessor.  After all, the
RiemannianTrustRegionSolver calls the assembler, so it may know the
function space basis.

[[Imported from SVN: r10035]]

[[Imported from SVN: r9989]]