Refactor interactions and evaluators module

[kilian]

... continuing the task of #706.

I've yet to decide whether state_matrix_t should become the base type for interactions and then evaluators, or state_matrix_t should be abstracted as a component. Note that there is also polarization_t.

[Juergen Reuter]

This is a hilariously big task, that clashes with all other work on the trunk. We should get this task done as soon as possible, or never change some of the old-fashioned data structures. Three of my attempts were not successful, so WK definitely has to take over here.

[Juergen Reuter]

This is a major blocker. It affects most of the code in many sections. All my attempts failed, so this will be moved.

[Juergen Reuter]

According to the discussions in Würzburg, only evaluator_t should inherit from interaction_t, but not interaction_t from state_matrix_t in order to allow for different implementations of state_matrix_t.

[Juergen Reuter]

Some transferral from global routines to TBPs happened, but caused paramount trouble with the gfortran 4.9 and 5.0. Most probably this is not gonna happen at all.

[Juergen Reuter]

As of r6795, evaluator_t has inherited from interaction_t, and all initializers of evaluator_t have been OO (there are no different routines distinguishing between interaction and evaluator types as inputs.). Almost all routines of interaction_t and evaluator_t have become TBPs.

[Juergen Reuter]

This has been done in r6798. There are still some places left for polishing interactions.f90 but this can be done once in a while. Closing.