Add Statistical Significance for Subgroup Discovery

[Rilanja]

This PR introduces a framework for assessing statistical significance of discovered subgroups via permutation testing, implemented through composition to preserve original functionality.

Key Changes:

1. StatisticalSignificance Class

   • Performs permutation tests to generate null distribution
   • Progress bar integration
   • Supports multiple testing corrections (Bonferroni, Holm, ...)
   • Implements both parametric (normal, gumbel_r) and non-parametric (empirical p-values)
   • Handles different target types (binary, numeric, frequent itemsets)

2. Stats

   • Wrapper for existing search strategies
   • Parallel permutation testing implementation

3. SignificantSubgroupResult

   • Enhanced result class with p-values (incl. method name in column header) and adjusted p-values
   • Adds statistical metrics to dataframe

4. Visualization

   • plot_null_distribution() function for analysis:
     • Histogram + KDE of null distribution
     • Observed quality reference line
     • Optional comparison with theoretical distributions (Gumbel/Normal)

Technical Notes:

• Required monkey-patching of SelectorBase serialization methods to enable parallel processing
• New dependencies:
  • joblib (parallel processing)
  • statsmodels (multiple testing correction; normal and gumbel distribution)
  • tqdm and tqdm_joblib (progress bar and patch for parallel processing)
  • seaborn (Plotting of histogram and KDE)
• Install all with:

pip install joblib statsmodels tqdm tqdm_joblib seaborn

Usage Example:

• Added metrics.ipynb demonstrating:
  • Significance testing workflow
  • Multiple comparison corrections
  • Null distribution visualization