Add bimodal_correlation_threshold to NumericImputationConfig

[DEVunderdog]

Parent

#105

What to build

Add bimodal_correlation_threshold: float = 0.2 to NumericImputationConfig in src/dataforge_ml/imputation/_config.py. This is the minimum absolute Pearson |r| a feature must have against a bimodal column for it to count toward the branch 2/3 feature tally in the Bimodal Imputation Framework.

The field is intentionally separate from mcar_feature_predictability_threshold — those two thresholds answer different questions: bimodal feature-counting asks "does this feature know which cluster a row belongs to"; MCAR predictability asks "can a model predict this column's value". They deserve independent knobs.

The hardcoded 0.2 at _numeric_imputer.py:652 (used when computing branch 3 feature centroids during fit) must be replaced with config.bimodal_correlation_threshold in the same PR.

Files

• src/dataforge_ml/imputation/_config.py — add field, docstring entry in NumericImputationConfig, to_dict(), from_dict()
• src/dataforge_ml/imputation/_numeric_imputer.py — replace hardcoded 0.2 at line 652 with config.bimodal_correlation_threshold
• tests/unit/imputation/test_imputation_config.py — round-trip test for the new field

Acceptance criteria

• NumericImputationConfig has bimodal_correlation_threshold: float = 0.2 with a numpy-style docstring entry (ADR-0034)
• to_dict() serialises the field; from_dict() deserialises it with the correct default
• The hardcoded 0.2 at _numeric_imputer.py:652 is replaced with config.bimodal_correlation_threshold
• Unit test confirms the field round-trips through to_dict() / from_dict() correctly
• All existing tests pass

Blocked by

None — can start immediately.