Major changes:
- borg() is now the unified entry point with two modes:
  - Diagnosis mode: detects dependencies and generates valid CV folds
  - Validation mode: validates existing train/test splits
- borg_diagnose(): automatic detection of spatial, temporal, and clustered dependencies
- borg_cv(): generates appropriate CV schemes (spatial_block, temporal_block, group_fold)
- Random CV blocked when dependencies detected (requires allow_random=TRUE to override)
- Inflation estimation for random CV bias
- New visualization functions: plot_split, plot_risk, plot_temporal, plot_spatial, plot_groups
- Extended model inspection support: lm, glm, ranger, xgboost, lightgbm, parsnip, workflows