Make install automatically moves the specified into the default share
directory, so the initially build library can go with the whole build dir

biggest change: make it rely on ost docker-container as base image

So sad to remove our beloved tetris. However, it somehow broke with Python3
and I did not have the nerves to fix it. Any resurection effort is very welcome.

- os.path.walk does not exist in Python3 anymore, use os.walk instead
  (changed interface)
- sphinx-build tried really hard to use Python 2. Import sphinx from
  current interpreter instead and directly call its __main__ function

As weird as it sounds but the proper var. to set is PNG_PNG_INCLUDE_DIR instead
of PNG_INCLUDE_DIR. The latter is overwritten by FindPNG and actually obsolete
(see https://github.com/Kitware/CMake/blob/master/Modules/FindPNG.cmake).

Several chain mappings are valid in this case. Only check whether we found an
automated mapping with strict (stuff must really match) chain_mapping_scheme

This only happens when running the unit tests due to increased log levels.
These warnings could manually be triggered for normal scripts with:
python -Wd world_domination.py.

Don't ask, just feels better that way

Requirement of Python 3 to fulfill our promise of being able to set
various things. This also removes an ugly hack of setting a variable
intended to be private in the unit test.

Due to different sorting behaviour in Python 2 and 3, the found symmetry
groups might differ while still being conceptually correct. Instead of
expecting exact results, the unit test now only checks for the number
of found groups and their size. Size is checked on the first element,
equal size of those elements is checked elsewhere.

import_array() must be called when the numpy C-API is called.
This is a macro that in our case defined a return value that makes
the compiler complain. The added #define statement enforces no return value

- Fix python code that is injected into python interpreter
- Proper error handling => fatal_error to stop cmake run if numpy not found

This can happen if python has been compiled with a specialized
malloc.