Implement DG method

[schmoelder]

Works by Kristian Meyer have shown speed improvements in the order of 10^2 using Discontinuous Galerkin method (standard nodal DG-method). This could be implemented for CADET.

[sleweke]

This would actually simplify a lot of the GRM, LRMP, and GRM2D code, because we can remove the domain decomposition inside the unit operation. We just dump the Jacobian into a large / big-ish sparse matrix and use a direct solver. We then revert to the adaptive Jacobian update strategy of IDAS.

Careful planning could also simplify implementation of the unit operations:

• Each operator can be abstracted into a function or class
• Models are simply composed by inserting the operators:

  dc_dt = -u * convection(c) + D_ax * dispersion(c) - ...
  

• This is basically an embedded DSL in the C++ code
• Use expression templates to generate fast element assembly code

Implementing DG shouldn't be that hard, I think.

[jbreue16]

DG method added in #110