fix: LagsAdder.features_added() over-reports lags outside the horizon window

[Valyrian-Code]

What

LagsAdder.transform() only materializes lags that fall within a horizon's valid window (>= the horizon and <= history_available) — a configured lag below the shortest horizon, or beyond the available history, is never added. But features_added() returned a name for every configured lag (self._lags), so it advertised columns that transform() never creates.

That is a contract mismatch: any consumer relying on features_added() (feature selection, column validation) is told about columns that never materialize. Same defect shape as the recently merged #965 (RollingAggregatesAdder) and #980 (DatetimeFeaturesAdder).

Concretely, with history_available=10d, horizons=[2D], custom_lags=[1D, 3D, 30D], add_trivial_lags=False:

• transform() adds only load_lag_P3D (1D is below the 2D horizon, 30D exceeds the 10D history);
• features_added() previously returned load_lag_P30D, load_lag_P3D, load_lag_P1D.

Fix

Return only the lags transform() actually produces — the union of the per-horizon valid lags (self._horizon_lags), which mirrors the columns built in _transform_single_horizon / _transform_versioned:

materialized_lags = sorted({lag for lags in self._horizon_lags.values() for lag in lags}, reverse=True)
return [self._lag_feature(lag) for lag in materialized_lags]

No change to transform(), the lags/horizon_lags properties, or any public field.

Tests

Added regression tests (both fail on the previous code, pass with the fix):

• single-horizon: features_added() equals the columns transform() adds, with the out-of-window custom_lags dropped;
• multi-horizon: features_added() equals the union of columns created across horizons.

ruff check, ruff format --check, and ty check are clean on the changed files. The openstef-models time-domain transforms suite (64) and the openstef-meta presets suite (29 — the one internal consumer, forecasting_workflow.py, uses this as an exclude set where the change is behaviour-preserving) both pass.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.