Changes in version 0.4.0                        

  - Added unified auditing for binary classification, multiclass
    classification, and regression.
  - Added automatic task detection with explicit overrides for numeric
    class codes.
  - Added multiclass confusion matrices, class-level
    precision/recall/F1, macro and weighted metrics, one-vs-rest AUC,
    multiclass log loss and Brier score, and top-label calibration.
  - Added regression MAE, RMSE, MSE, R-squared, bias, MAPE, SMAPE,
    normalized errors, residual diagnostics, and subgroup error
    analysis.
  - Extended pass/fail requirements and plain-language risk warnings to
    all supported tasks.
  - Updated HTML, Markdown, JSON, and plotting behavior to be
    task-specific.