Optimize Python Interface for Performance

[Bonifatius94]

Issue:

• the costs of calling the Python C-API seem to outweigh all time savings of implementing chess functionality in C
• therefore it might not make a lot of sense to expose the C code to Python at this fine-grained level anymore

Suggested Solution:

• re-think the API interface such that it still makes sense to expose the lib to Python
• consider options for generating larger training batches, e.g. play whole games with a single function call, add preprocessing to retain valid training batches right away
• even think of implementing the whole AI in C/C++ via the TensorFlow C/C++ API