fix(ci): support scikit-learn 1.9 (LassoCV n_alphas removal)

scikit-learn 1.7 deprecated LassoCV's `n_alphas` in favour of an integer
`alphas`, and 1.9 removed it — `LassoCV(n_alphas=20)` now raises TypeError
at construction. The CI matrix resolves to scikit-learn 1.9.0, breaking
`sp.tmle(method='hal')` (HALRegressor) and `sp.rd_flex(learner='lasso')`.

Add a version-robust shim `compat.sklearn.lasso_cv_alphas_kwargs(n)` that
emits {"alphas": n} on sklearn >= 1.7 and {"n_alphas": n} on older
releases; route both call sites through it. Alpha count unchanged (20 / 50)
— no numerical effect.

Verified: 8 HAL tests + 6 rd_flex tests green on local sklearn 1.6.1;
version-logic asserted across 1.6/1.7/1.8/1.9; flake8 hard + baseline and
mypy baseline gates pass locally.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: brycewang-stanford

CHANGELOG.md

@@ -23,6 +23,22 @@ All notable changes to StatsPAI will be documented in this file.
 
 ### Fixed
 
+- **CI: scikit-learn 1.9 compatibility — `LassoCV(n_alphas=...)` removal.**
+  scikit-learn 1.7 deprecated the `n_alphas` argument of the coordinate-descent
+  CV estimators (`LassoCV`/`ElasticNetCV`/…) in favour of passing an integer to
+  `alphas`, and 1.9 removed it outright — constructing `LassoCV(n_alphas=20)`
+  now raises `TypeError: LassoCV.__init__() got an unexpected keyword argument
+  'n_alphas'`. The `CI/CD Pipeline` matrix resolves to scikit-learn 1.9.0, so
+  `sp.tmle(method='hal')` (`HALRegressor`) and `sp.rd_flex(learner='lasso')`
+  both failed at construction (`tests/test_hal_tmle.py`,
+  `tests/test_estimator_provenance_round5.py::TestHalTmleProvenance`,
+  `tests/test_low_cov_battery.py::test_hal_regressor_predicts_finite`). A new
+  version-robust shim `statspai.compat.sklearn.lasso_cv_alphas_kwargs(n)` emits
+  `{"alphas": n}` on scikit-learn >= 1.7 and `{"n_alphas": n}` on older
+  releases; both call sites now route through it. The number of path alphas
+  (20 for HAL, 50 for `rd_flex`) is unchanged — no numerical effect. Verified
+  on the local scikit-learn 1.6.1 pin (8 HAL tests + 6 `rd_flex` tests green)
+  and by version-logic assertion across 1.6/1.7/1.8/1.9.
 - **CI: `schemas/functions.json` no longer drifts by pandas version.** The
   auto-registered schema export stringified parameter annotations via
   `str(typing.Optional[pandas.DataFrame])`, which pandas 3.0 renders as

src/statspai/compat/sklearn.py

@@ -41,6 +41,35 @@ def _check_sklearn():
         )
 
 
+def lasso_cv_alphas_kwargs(n_alphas: int) -> dict:
+    """Version-robust keyword for the number of CV path alphas.
+
+    scikit-learn 1.7 deprecated the ``n_alphas`` argument of the
+    coordinate-descent CV estimators (``LassoCV`` / ``ElasticNetCV`` / ...)
+    in favour of passing an integer to ``alphas``; ``n_alphas`` was removed
+    outright in 1.9 (raising ``TypeError`` on construction). Older versions
+    only understand ``n_alphas``. Return whichever keyword the installed
+    scikit-learn accepts so call sites stay portable across the boundary.
+
+    Parameters
+    ----------
+    n_alphas : int
+        Number of alphas to evaluate along the regularisation path.
+
+    Returns
+    -------
+    dict
+        ``{"alphas": n}`` on scikit-learn >= 1.7, else ``{"n_alphas": n}``.
+    """
+    from packaging.version import parse as _parse
+
+    import sklearn
+
+    if _parse(sklearn.__version__).release[:2] >= (1, 7):
+        return {"alphas": int(n_alphas)}
+    return {"n_alphas": int(n_alphas)}
+
+
 class SklearnOLS(BaseEstimator, RegressorMixin):
     """
     Sklearn-compatible OLS with robust/clustered standard errors.

src/statspai/rd/rd_flex.py

@@ -287,8 +287,9 @@ def _make_learner(name: str, random_state: Optional[int]):
             return RidgeCV(alphas=np.logspace(-3, 3, 21))
         if name == "lasso":
             from sklearn.linear_model import LassoCV
-            return LassoCV(cv=5, n_alphas=50, random_state=random_state,
-                           max_iter=10_000)
+            from ..compat.sklearn import lasso_cv_alphas_kwargs
+            return LassoCV(cv=5, random_state=random_state,
+                           max_iter=10_000, **lasso_cv_alphas_kwargs(50))
     except ImportError as exc:  # pragma: no cover
         raise ImportError(
             "rd_flex requires scikit-learn. Install with: pip install scikit-learn"

src/statspai/tmle/hal_tmle.py

@@ -176,11 +176,12 @@ def fit(self, X, y):
         B, anchors = _hal_basis(X, anchors=None,
                                  max_anchors_per_col=self.max_anchors_per_col)
         from sklearn.linear_model import Lasso, LassoCV
+        from ..compat.sklearn import lasso_cv_alphas_kwargs
         if self.lambda_ is None:
             cv = int(max(2, min(self.cv, max(2, len(y) // 20))))
             model = LassoCV(
                 cv=cv, random_state=self.random_state,
-                n_alphas=20, max_iter=5000,
+                max_iter=5000, **lasso_cv_alphas_kwargs(20),
             )
         else:
             model = Lasso(alpha=self.lambda_, max_iter=5000,