feat: add Bayesian model criticism example and guide

[jc-macdonald]

Summary

Adds a third user-guide example demonstrating the package's most distinctive features: proper scoring rules, calibration assessment, and score-based model averaging (#71).

New files

• examples/bayesian_study.py — runnable script with snippet tags
• docs/guide/bayesian.md — narrative walkthrough referencing snippets
• docs/assets/bayesian_*.png — generated plot assets

Features demonstrated

Feature	API used
Proper scoring rules	score("crps", ...), score("coverage", ...), score("rmse", ...)
Calibration assessment	coverage_curve() + plot_calibration()
Score-based stacking	stack_scores() + ensemble_predict()
Factor screening	screen(..., method="morris") + reduce_factors()
Annotations	Annotation(name, lookup, key) for compute cost
Persistence	save_results() / load_results() round-trip
Pareto analysis	extract_front() with weighted observables
Plotting	plot_front(), plot_parallel(), plot_scores()

Design

Conjugate Bayesian linear regression (y = a + bx + eps) with Normal prior — closed-form posterior, no MCMC, no external probabilistic programming library. Three factors control prior variance, assumed noise scale, and sample size.

CI

• 258 tests pass, 99.57% coverage (no new tests needed — this is a docs/example PR)
• Ruff, mypy all clean
• Example script runs end-to-end and produces all 4 plot assets

Closes #71