seperated assembling of global stiffness matrix in three essential parts:

1) Generation of sparse matrix
2) Assembling of global stiffness matrix from local stiffness matrices
3) Interface for providing a local stiffness matrix and boundary conditions

All this is only for P1, but it works with hanging nodes, in parallel (with
algebraic extension of nonoverlapping grids) and for systems.
(Systems are yet untested).

[[Imported from SVN: r3653]]