Add performence tests / performance awareness

[Bonifatius94]

Issue:

• performance is currently not properly measured
• if the performance is unknown, it's really hard to make improvements
• high performance is the key feature of this project (!!!), needs to perform better than naive implementations

Suggested Solution:

• think of useful ways to measure the performance (e.g. code profiling tools showing elapsed times of each function call during a high-performance test)
• think of performance constraints when releasing the package (critical functions may not take significantly longer than on previous releases)

[Bonifatius94]

The performance issues might relate to lots of Python C-API calls which seem to be rather slow. Therefore think of ways to save API calls:

• just expose a function predict_best_draw(chessboard), keep the rest local
• add a transposition table for caching already evaluated positions
• implement the AI part with TensorFlow in C/C++ rather than in Python