Closes #4, superseeds #8.

@cortner A first version, which seems to work and produce the right results. Convergence is still pretty slow, though.

• Tests for DofManager
• Rethink the integration if this can be done better
• Use Voigt convention to track the 6 degrees of freedom in the lattice deformation matrix.
• Make the mapping of (position, lattice vectors) to degrees of freedom more flexible, such that custom constraints or different ways of parametrising the cell could be easily tried ... maybe only after this PR.