v1.1.0

• Feature: Confidence intervals for action rules via ActionRules.confidence_intervals() with three engines — bootstrap (percentile, BCa), analytic (Wald, Newcombe-Wilson, auto), and Bayesian (Beta-Binomial Monte Carlo).
• Feature: Rule categorisation (accept / reject / uncertain) based on a user-supplied decision threshold over uplift or realistic_rule_gain.
• Feature: Cross-validation via ActionRules.cross_validate() with stratified folds, configurable selection strategies, and out-of-fold metric estimation.
• Feature: New action_rules.visualization module with forest plots, bootstrap distributions, coverage calibration, and CI-width diagnostics (matplotlib lazy-imported).
• Feature: CI results are surfaced in Output.get_ar_notation(), Output.get_pretty_ar_notation(), and JSON export (get_export_notation()); NaN/Inf are serialised as null.
• Feature: New CLI options for confidence intervals (--ci-method, --ci-level, --ci-threshold, --ci-metric, …).
• Fix: df_to_array no longer creates spurious <attr>_<item_*>_nan one-hot columns for missing antecedent values; NaN is now preserved through get_dummies per the pessimistic null-value semantics of Dardzinska (2013, §2.3.2). Target column behaviour unchanged.
• Docs: Notebooks for the Telco churn end-to-end tour, rule-level CIs across three datasets (Bank Marketing, Credit Card Default, Employee Attrition), and the inference studies / article figures.