Create basic self-play algorithm

[TTitcombe]

Create a basic algorithm which takes the board as input and predicts a move.
See how well this model can learn playing against itself, using sparse rewards (win or lose) and frequent rewards (take or lose a piece)