This is done by solving a mass matrix system.

[[Imported from SVN: r6891]]

[[Imported from SVN: r6890]]

[[Imported from SVN: r6889]]

[[Imported from SVN: r6888]]

add a scaled identity to the rod stiffness matrix if there is no Dirichlet boundary: this regularizes the problem

[[Imported from SVN: r6886]]

[[Imported from SVN: r6884]]

[[Imported from SVN: r6883]]

[[Imported from SVN: r6882]]

[[Imported from SVN: r6879]]

Finish implementation of SteklovPoincareStep with contact problems.  I don't dare to test this today...

[[Imported from SVN: r6871]]

[[Imported from SVN: r6870]]

This handles the case that there are contact/mortar relations
between the continua.  Unfortunately, this does not quite fit
into the framework: so far we have treated all continua individually.
However, the contact solvers lumps everything into a single
algebraic system.  Consequently this is now getting a bit
hacky.  I hope to find something prettier in the future.

The linearized NtD map for continua is not implemented yet.

[[Imported from SVN: r6868]]

[[Imported from SVN: r6866]]

[[Imported from SVN: r6863]]

[[Imported from SVN: r6862]]

[[Imported from SVN: r6858]]

[[Imported from SVN: r6857]]

This completes the implementation.  In principle, the Steklov-Poincare step
should now work for arbitrary configurations of rods and continua.

Known issues:
- rods must not stick to the same continuum with both ends
- all objects must have a nonempty Dirichlet boundary
  (will be fixed very soon)

[[Imported from SVN: r6856]]

[[Imported from SVN: r6855]]

[[Imported from SVN: r6854]]

[[Imported from SVN: r6853]]

[[Imported from SVN: r6852]]

[[Imported from SVN: r6851]]

[[Imported from SVN: r6850]]

[[Imported from SVN: r6849]]

[[Imported from SVN: r6848]]

[[Imported from SVN: r6847]]

[[Imported from SVN: r6846]]

SteklovPoincareStep class.

That way it can access all its data.  I don't know if that is better
design-wise, but for the time being it make things simpler.

[[Imported from SVN: r6845]]

The entire primal (i.e., Dirichlet-to-Neumann) should now run for
arbitrary numbers of rods and continua.

[[Imported from SVN: r6844]]

[[Imported from SVN: r6843]]

[[Imported from SVN: r6842]]

[[Imported from SVN: r6841]]

Because that is what happens.  The previous solution of returning
to FieldVectors, one as the return value and the other one using
call-by-reference, was extremely ugly.

This is an API change.

[[Imported from SVN: r6840]]

[[Imported from SVN: r6839]]

[[Imported from SVN: r6838]]

[[Imported from SVN: r6837]]

[[Imported from SVN: r6836]]

[[Imported from SVN: r6835]]

Create rod by interpolating between two endpoints, which are expected to be already in the result container

[[Imported from SVN: r6834]]