CI runners seem to stop working half way and still report successful execution

Some tests that are known to be KO are still reported successful.

For instance:

Expected result should look like: https://gitlab.orfeo-toolbox.org/s1-tiling/s1tiling/-/jobs/44035

We see that pytest seems to halt half way and the job still reports everything is fine, but we miss the final report summary:

---------- generated xml file: /builds/s1-tiling/s1tiling/report.xml -----------
=========================== short test summary info ============================
FAILED tests/test_0200306-NR.py::test_33NWB_202001_NR[baselinedir0-srtmdir0-False-outputdir0-tmpdir0-False]
=================== 1 failed, 1 passed in 691.43s (0:11:31) ====================

In some other tests, it seems nothing has been run:

Actually we can see the log halts in middle on pip execution

Building wheel for S1Tiling (setup.py): started

the expected following

Building wheel for S1Tiling (setup.py): finished with status 'done'

is nowhere to be found.

After further investigation it may be related to a communication issue between the job and the CI when kubernete is used. See the related issues:

which final resolution seems to be currently tracked under:

Should we make sure to spend our time logging? (https://gitlab.com/gitlab-org/gitlab-runner/-/issues/26654) and/or should we use the new FF_USE_LEGACY_KUBERNETES_EXECUTION_STRATEGY=false option?