v0.1.0

Initial release of Bayesian BM25.

• Sigmoid likelihood + composite prior + Bayesian posterior framework for converting raw BM25 scores into calibrated probabilities
• Batch parameter learning via gradient descent and online learning via SGD with EMA-smoothed gradients and Polyak averaging
• Probabilistic score fusion: prob_and, prob_or, and log_odds_conjunction
• Integrated search scorer wrapping bm25s (BayesianBM25Scorer)

Install: pip install bayesian-bm25