Fix low level crash in BisectingKMeans.predict on rescaled data - #27167

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: a4a5abf. Link to the linter CI: here}

Victory! The non-regression test is actually causing a crash on the macOS intel CI as expected:

https://dev.azure.com/scikit-learn/scikit-learn/_build/results?buildId=58341&view=logs&j=97641769-79fb-5590-9088-a30ce9b850b9&t=4745baa1-36b5-56c8-9a8e-6480742db1a6

Current thread 0x0000000107bdae00 (most recent call first):
  File "/Users/runner/work/1/s/sklearn/cluster/_kmeans.py", line 796 in _labels_inertia
  File "/Users/runner/work/1/s/sklearn/cluster/_kmeans.py", line 820 in _labels_inertia_threadpool_limit
  File "/Users/runner/work/1/s/sklearn/cluster/_bisect_k_means.py", line 504 in _predict_recursive
  File "/Users/runner/work/1/s/sklearn/cluster/_bisect_k_means.py", line 520 in _predict_recursive
  File "/Users/runner/work/1/s/sklearn/cluster/_bisect_k_means.py", line 471 in predict
  File "/Users/runner/work/1/s/sklearn/cluster/tests/test_bisect_k_means.py", line 148 in test_non_regression_segfault_on_empty_bisections

Let's now check that the fix works as expected.

Note: this PR implements the fix both in the Python caller and the Cython callee:

• sklearn/cluster/_bisect_k_means.py
• sklearn/cluster/_k_means_lloyd.pyx

In theory, only one of those two fixes is actually needed, but I am not sure where it's best to fix this bug so I decided to fix it at both locations.

EDIT: I changed my mind and decided to make to make it explicit in the Cython code that it's expected to call the no-op case. This makes the Python code of bisecting k-means simpler as a result.

The tests pass! I will generalize the fix to Elkan next week as suggested by @jjerphan. Have a nice WE.

In the end I decided to only keep the Cython side of the fix for the sake of simplicity. I updated the comments accordingly.