Flip strict=False -> strict=True on zips with invariant-equal lengths

The ruff B905 auto-fix inserted `strict=False` everywhere to preserve
pre-existing behavior. Audit of each site showed the conservative choice
was wrong for most of them: the two iterables are paired by construction
or invariant (feature_names to columns, estimators to weights,
pred_np rows to y_indices, etc.), so truncation would hide a bug rather
than be desired behavior.

The only genuine intentional-truncation case is meta_models.py's
`zip(range(n_estimators), relevant_config_product)`, where
`n_estimators_per_model = max(n_estimators // len(product), 1)`
deliberately allows the two lengths to differ. That one stays
`strict=False`.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: adrian-prior

src/tabpfn_extensions/sklearn_ensembles/weighted_ensemble.py

@@ -30,7 +30,7 @@ def fit(self, X, y):
         # Prune classifiers with weights below the threshold
         pruned_classifiers = []
         pruned_weights = []
-        for clf, weight in zip(self.estimators, weights, strict=False):
+        for clf, weight in zip(self.estimators, weights, strict=True):
             if weight >= self.weight_threshold:
                 pruned_classifiers.append(clf)
                 pruned_weights.append(weight)
@@ -40,7 +40,7 @@ def fit(self, X, y):
             pruned_classifiers = [
                 clf
                 for _, clf in sorted(
-                    zip(pruned_weights, pruned_classifiers, strict=False),
+                    zip(pruned_weights, pruned_classifiers, strict=True),
                     key=lambda pair: pair[0],
                 )
             ]

src/tabpfn_extensions/unsupervised/experiments.py

@@ -141,7 +141,7 @@ def run(self, tabpfn, **kwargs):
                         zip(
                             self.feature_names,
                             [self.X[:, i] for i in range(self.X.shape[1])],
-                            strict=False,
+                            strict=True,
                         ),
                     ),
                     "real_or_synthetic": "Actual samples",
@@ -156,7 +156,7 @@ def run(self, tabpfn, **kwargs):
                                 self.synthetic_X[:, i]
                                 for i in range(self.synthetic_X.shape[1])
                             ],
-                            strict=False,
+                            strict=True,
                         ),
                     ),
                     "real_or_synthetic": "Generated samples",

src/tabpfn_extensions/unsupervised/unsupervised.py

@@ -624,7 +624,7 @@ def outliers_single_permutation_(
                 if valid_indices.any():
                     # Get probabilities for each sample based on its class in y_predict
                     for idx, (prob_row, y_idx) in enumerate(
-                        zip(pred_np, y_indices, strict=False)
+                        zip(pred_np, y_indices, strict=True)
                     ):
                         if (
                             0 <= y_idx < pred_np.shape[1]

src/tabpfn_extensions/utils.py

@@ -444,7 +444,7 @@ def product_dict(d: dict[str, list[T]]) -> Iterator[dict[str, T]]:
     keys = d.keys()
     values = [d[key] for key in keys]
     for combination in itertools.product(*values):
-        yield dict(zip(keys, combination, strict=False))
+        yield dict(zip(keys, combination, strict=True))
 
 
 # Get the TabPFN models with our wrappers applied

tests/test_many_class_classifier.py

@@ -286,7 +286,7 @@ def predict_proba(self, X):
         for labels in fit_y_records:
             assert rest_code not in labels
 
-        for weights, labels in zip(fit_weight_records, fit_y_records, strict=False):
+        for weights, labels in zip(fit_weight_records, fit_y_records, strict=True):
             if weights is not None:
                 assert weights.shape[0] == labels.shape[0]