Rewrite FEValues/FiniteElement/Mapping interfaces

[bangerth]

Fixing #1152, I realized again that I do not understand (and have never really understood) the way FEValues, FiniteElement and Mapping classes interact. Looking at how all of this works I find that much isn't well documented, some of the interfaces could be improved, etc.

This issue is simply a catch-all for all of the patches I'm doing to improve the situation.

[guidokanschat]

I do not believe we should invest heavily in FEValues anymore. It is inherently memory bound and does not support tensor product integration like it is used in MatrixFree. Let's rather focus on replacing it and while we are doing so, separate Mapping in a way consistent with manifold descriptions.

[bangerth]

Fortunately, ours is a project big enough so that different people can work on different parts :-) (I'm not implementing new functionality, just cleaning up interfaces.)

[bangerth]

This is as far as I intend to take it for now.