EDA Toolkit 0.0.25

Fix: correct p-value precision loss in Bonferroni/BH adjustment

• Previously, p-values were rounded before being fed into the multiple comparison correction block, causing small but meaningful errors in adjusted values. Raw p-values are now stored as _raw_pval internally, the correction block reads from those, and a cleanup pass removes the key before DataFrame construction. All rounding is handled consistently via decimal_places at the formatting stage.

Added new normality_tests() function to eda_toolkit.

New Features

• Batch normality testing across numeric columns via Shapiro-Wilk,
  D'Agostino K², and Anderson-Darling
• alpha, tests, features, and decimal_places parameters
• Shapiro-Wilk auto-skipped for n > 5,000 with printed explanation
• Anderson-Darling uses method="interpolate" on scipy >= 1.17
  for real p-values; falls back to critical value comparison on
  older scipy
• Returns tidy summary DataFrame with Variable, Test,
  Statistic, P-value, Normal columns
• 40 pytest tests added