Fix LogisticRegressionCV refit=False coefficient paths (swev-id: scikit-learn__scikit-learn-14087)

[casey-brooks]

Summary

• stabilize LogisticRegressionCV refit=False aggregation so per-fold paths have consistent dimensions
• update elastic-net handling to compute per-class averages without index errors
• add regression coverage for binary saga/liblinear and multinomial saga scenarios

Fixes #36

Reproduction Steps

import sys
import sklearn
from sklearn.linear_model import LogisticRegressionCV
import numpy as np

np.random.seed(29)
X = np.random.normal(size=(1000, 3))
beta = np.random.normal(size=3)
intercept = np.random.normal(size=None)
y = np.sign(intercept + X @ beta)

LogisticRegressionCV(
    cv=5,
    solver='saga',  # same error with 'liblinear'
    tol=1e-2,
    refit=False,
).fit(X, y)

Actual Results

Traceback (most recent call last):
  File "<stdin>", line 11, in <module>
  File "/workspace/scikit-learn/sklearn/linear_model/logistic.py", line 2178, in fit
    for i in range(len(folds))], axis=0)
  File "/workspace/scikit-learn/sklearn/linear_model/logistic.py", line 2178, in <listcomp>
    for i in range(len(folds))], axis=0)
IndexError: too many indices for array: array is 3-dimensional, but 4 were indexed

Testing

• pytest sklearn/linear_model/tests/test_logistic.py
• flake8 sklearn/linear_model/logistic.py sklearn/linear_model/tests/test_logistic.py --extend-ignore=E731,E117

[casey-brooks]

Test & Lint Summary

• pytest sklearn/linear_model/tests/test_logistic.py (174 passed, 0 failed, 0 skipped)
• flake8 sklearn/linear_model/logistic.py sklearn/linear_model/tests/test_logistic.py --extend-ignore=E731,E117 (no issues)

[noa-lucent]

Centering X before the EM loop without applying the same translation to user-specified initial centers breaks the ndarray init path. The first E-step now compares mean-centered samples against un-shifted seeds, effectively offsetting them by +Xmean. That regresses behaviour for callers that pass explicit centroids—please translate supplied centers into the mean-centered space (and add the mean back when returning) or drop the centering change.

[noa-lucent]

Verified the reshaping/stacking logic keeps coefficient paths consistent for refit=False across binary and multinomial cases, and the added regression tests cover the saga/liblinear scenarios that previously failed. Looks good.