Improved correlation checks and reduced false positives in missing patterns

Improvements

Refined correlation checks in calculate_correlations
- Fixed type inference errors by iterating over analyzer.column_types instead of analyzer.df
- Updated mixed-variable thresholds to {'warning': 0.5, 'critical': 0.8} for consistency with Cramer’s V
- Ensured seamless integration with run_checks
Reduced over-flagging in missing patterns detection
- Introduced effect size thresholds:
  - Categorical: Cramer’s V > 0.1
  - Numeric: Cohen’s d > 0.2
- Tightened p-value threshold to 0.01
- Increased minimum samples per group to 10
- Replaced ANOVA (f_oneway) with Mann-Whitney U test for better handling of skewed distributions
- Added pattern grouping to summarize correlations per missing column (top 3 shown for conciseness)

Corrected correlation dictionary iteration (analyzer.column_types)
Prevented spurious warnings by filtering weak associations