Gomoku with AlphaZero style MCTS

Looking into the MCTS techniques used by Deepmind in AlphaZero and some potentially cool constraint solving stuff to add in.

Run python gomoku.py to start up the game. Edit GRID_LEN in constants.py to change the game type between 6x6 and 8x8.

pure_mcts.py and alphazero_mcts.py are the main files providing game-independent monte carlo tree search (can be applied to any game implementing the State interface in state.py). Currently using the pretrained models and the network architecture from https://github.com/junxiaosong/AlphaZero_Gomoku/blob/master/policy_value_net_numpy.py as the alpha zero style value network since the details seem complicated. Will take a further look into the architecture to fully understand it and see if improvements can be made.