Bayesian Network Inference Engine

A complete probabilistic inference engine for discrete Bayesian networks. It supports exact inference by enumeration over the topological ordering, and three approximate sampling methods when exact inference gets too expensive. The Gibbs sampler is the interesting one, because it needs the Markov blanket of each variable to resample it correctly, and that is computed from the graph rather than hardcoded.

Attention: Coursework, kept because it is a genuinely complete implementation rather than a skeleton. It has its own test suite.