FIX KMeans with fortran-aligned data.

bad case:
X = np.asfortranarray([[0, 0], [0, 1], [0, 1]])
labels = np.array([0, 1, 1])
km = KMeans(n_init=1, init=centers, precompute_distances=False)
km.fit(X)
print km.cluster_centers_
print km.labels_
print km.n_iter_

Author: zhai_pro