v0.5.9

Patch release improving custom evaluation split handling and preserving FeatureValidation metadata during CreditModel handoff.

Custom Evaluation Splits

• ScoreComparisonPipeline.split_col now accepts any non-empty evaluation label, including values such as ft_oot, instead of limiting users to ins, oos, and oot.
• FeatureValidationPipeline still requires ins and oos for fitting and validation, but preserves additional split labels as evaluation datasets.
• Additional evaluation splits are transformed with the fitted WOE engine but never participate in WOE fitting. CSV feature-batch mode preserves the same labels and behavior.

FVP to CreditModel Handoff

• FeatureValidationPipelineResult now retains a serializable configuration snapshot even when selection_enabled=False.
• FeatureScreeningArtifact.from_fvp_result() recovers target, feature, WOE, and weight_col metadata from that snapshot, allowing direct reuse by CreditModelPipeline without a false weight mismatch.
• WOE fitting remains unweighted; weight_col metadata applies to weighted screening, model training, and evaluation stages.

Validation

• Full local pytest suite: 632 passed.
• Python 3.11/3.12 legacy, modern, and bleeding CI matrices passed before the release bump.
• Source package verification and mkdocs build --strict passed.