For this, dlopen needs to share symbols globally because MKL internally
uses dynamic loading that doesn't work otherwise.

Add parts to configure.ac and Makefile.am. There is now an additional
flag --enable-python-tests. The tests require the pytest package.

The implementation of the test is a bit hacky because it takes
wrapper.so from .libs. This should probably improbed for better
portability.

This is more efficient because the python loop processes only (number of
blocks)^2 elements.

By default, python wrapper is not built, only if this switch is
specified.

- depends on python, cython, mpi4py and numpy
- currently only dependency check on python
- should be done optionally
- converts wrapper.pyx to wrapper.c using cython, then compiles this to
  a python extension
- installs the files such that they can be imported with `import pyelpa`
  in python

The GPU initialization is actually quite constly, e.g. on Minsky it
takes roughly 0.7s. That is hurting performance for small matrices.
Thus a check has been added and now GPU should be initialized only the
first time.