Implement missing 4th-edition algorithms (game theory, EM, Kalman, DBN, DDN, SARSA) with tests and demo notebooks, plus bug fixes and CI modernization

[dmeoli]

This PR fills in several algorithms from the 4th edition that were not yet implemented, adds tests (including cases taken from the book and the 3rd-edition solutions manual) and runnable demo notebooks, fixes a few bugs, and modernizes the CI. The full test suite passes (446 tests).

New algorithms

game_theory.py (new module) - Chapter 18, Multiagent Decision Making

• 18.2 Non-cooperative: dominates / dominant_strategy, iterated_dominance, pure_nash_equilibria, solve_zero_sum_game (von Neumann minimax via linear programming)
• 18.3 Cooperative: shapley_value, is_in_core
• 18.4 Collective decisions: plurality_winner, borda_winner, condorcet_winner, vickrey_auction, contract_net, alternating_offers_bargaining

probability.py

• KalmanFilter / kalman_filter - Kalman filtering (Section 15.4)
• DynamicBayesNet - dynamic Bayesian networks with unroll() and exact filtering (Section 15.5)
• baum_welch - EM for learning HMM parameters (Section 20.3.3)

learning.py

• gaussian_mixture_em - EM for a mixture of Gaussians (Section 20.3.1)
• naive_bayes_em - EM for a Bayes net with a hidden variable, the candy-bags model (Section 20.3.2)

mdp4e.py

• pomdp_lookahead and update_belief - online POMDP agent over a dynamic decision network via belief-state expectimax look-ahead (Section 17.4)

reinforcement_learning.py and reinforcement_learning4e.py

• SARSALearningAgent - on-policy temporal-difference control (Section 21.3)

Bug fixes

• return NotImplementedError changed to raise NotImplementedError in deep_learning4e.py and ipyviews.py (was returning the exception class instead of raising it)
• Fixed SyntaxWarning from invalid escape sequences in docstrings (csp.py, logic.py) by using raw strings
• Keras 3 compatibility: optimizers.SGD(lr=...) changed to learning_rate=... in deep_learning4e.py
• Made test_value_iteration robust to floating-point rounding across numpy/BLAS versions (pytest.approx)

Tests

• New tests/test_game_theory.py and new tests for the Kalman filter, DBN, Baum-Welch, both EM variants, the DDN look-ahead agent, and SARSA
• Cases taken from the book / 3rd-edition solutions manual: the zero-sum game of exercise 17.17 (optimal mixed strategy [1/9, 1/9, 1/9, 1/3, 1/3], value 0), a saddle-point game, battle of the sexes, stag hunt, the umbrella DBN cross-checked against the HMM forward algorithm, and the Kalman
  steady-state variance

Demo notebooks

• Added executed notebooks with output (and plots): game_theory.ipynb, expectation_maximization.ipynb, kalman_filter.ipynb, dynamic_decision_network.ipynb, sarsa.ipynb

Cleanup

• Fixed spelling typos in comments, docstrings, and data labels (codespell)
• Updated the README algorithm index for all the new entries and corrected the Python version section (now Python 3.9 and up)

CI

• Added a GitHub Actions workflow running the full suite on Python 3.9, 3.10, 3.11, and 3.12
• Bumped the Python versions in .travis.yml to match (3.9-3.12)
• Updated the build-status badge

[dmeoli]

@antmarakis @norvig