Refactor and merge Hans' and Ulli's polygon support

[ukoethe]

[hmeine]

Concerning my polygon class, I wonder if a cleaner approach could be a combination of

• a data container class based on NumpyArray[View](thus offering full numpy compatibility and potential) instead of std::vector as it is now, and
• free functions containing the different algorithms.

The implementation I have is based on a PointArray class (originally intended to represent a set, not necessarily even a sequence of points), which wraps a std::vector, and a derived Polygon class which has additional conceptual properties (e.g. arc length, which is not defined on a point array), data members caching various interesting values, and some member functions.

This has the following problems:

• The PointArray base class restricts the use of the algorithms to other point array representations, i.e. we cannot work with views of e.g. data from numpy arrays.
• The "is-a" relationship between the two classes is broken, due to the caching data members. I.e. if I use PointArray's modifying methods, the meta information managed by the derived Polygon class becomes invalid.

I think the design needs to be revamped, while the algorithms should be quite easy to reuse, and the existing functionality gives a good impression of what's commonly needed.