WIP Bugfix/petsc operator

Try to fix all kinds of problems with the petsc bindings:

1. with continuous functions and non zero initial guess, petsc inverse operator produces wrong values on the border (Lagrange space)
2. petsc matrices and vectors have states (add, insert) and we almost totally ignore that - leads to problems
3. at the moment with lagrange space the example-solver with petsc doesn't work with higher number of processors
4. other stuff...