Use _openmp_effective_n_threads to take cgroups CPU quotas into account into account in Elkan and Minibatch KMeans #20483
Description
A partial output of git grep prange shows that those Cython file do not always path an explicit num_threads which means that those parallel loops will typically be over-subscribed in a docker container with low CPU quotas:
sklearn/cluster/_k_means_elkan.pyx:from cython.parallel import prange, parallel
sklearn/cluster/_k_means_elkan.pyx: for i in prange(n_samples, schedule='static', nogil=True):
sklearn/cluster/_k_means_elkan.pyx: for i in prange(n_samples, schedule='static', nogil=True):
sklearn/cluster/_k_means_elkan.pyx: for chunk_idx in prange(n_chunks, schedule='static'):
sklearn/cluster/_k_means_elkan.pyx: for chunk_idx in prange(n_chunks, schedule='static'):
sklearn/cluster/_k_means_lloyd.pyx:from cython.parallel import prange, parallel
sklearn/cluster/_k_means_lloyd.pyx: for chunk_idx in prange(n_chunks, schedule='static'):
sklearn/cluster/_k_means_lloyd.pyx: for chunk_idx in prange(n_chunks, schedule='static'):
sklearn/cluster/_k_means_minibatch.pyx:from cython.parallel cimport parallel, prange
sklearn/cluster/_k_means_minibatch.pyx: for cluster_idx in prange(n_clusters, schedule="static"):
sklearn/cluster/_k_means_minibatch.pyx: for cluster_idx in prange(n_clusters, schedule="static"):
Similar to #20477.