Add Dots & Boxes engine and AlphaZero self-play loop

[krandder]

Motivation

• Provide a reusable, configurable engine for the Dots and Boxes game that supports arbitrary (rows, cols) board sizes and game-state serialization.
• Add an AlphaZero-style self-play pipeline to generate training data via MCTS for experimenting with policy/value networks.
• Expose a small public API so other modules can import the game, MCTS, and self-play utilities easily.

Description

• Add dots/engine.py implementing DotsAndBoxesConfig, DotsAndBoxesState, and DotsAndBoxes with action mapping, move application, scoring, serialize, legal_actions, render_edges, legal_action_mask, and normalize_policy utilities.
• Add dots/alphazero.py implementing MCTSConfig, MCTSNode, MCTS, PolicyValueNet interface and a UniformPolicyValueNet baseline, plus select_action and self_play which produces TrainingExample records from self-play using visit-count policies and Dirichlet noise at the root.
• Add package initializer dots/__init__.py to re-export DotsAndBoxes, DotsAndBoxesState, MCTS, MCTSConfig, PolicyValueNet, UniformPolicyValueNet, TrainingExample, and self_play for convenient imports.
• MCTS uses configurable num_simulations, c_puct, and root Dirichlet noise blending, and self_play records per-state policy vectors and final game outcome values.

Testing

• No automated tests were executed as part of this change (no test suite was run).

---

Codex Task