[PierreRaybaut]

After #402 was merged, the PyQt5 / Python 3.9 job on develop crashed twice with a segmentation fault (exit code 139) at the start of side_by_side_app_test.py::test_view_images_side_by_side (run 39420412512, both attempts), while the same tree had passed on the PR branch with identical dependencies. The logs contained no traceback at all, so the crash could not be analyzed.

Why no traceback was reported

The application context (sigimax_app_context) redirects faulthandler to .DataLab_faulthandler.log in the configuration directory, then disables it on exit, which also disables pytest's own handler for every subsequent test. Crash tracebacks therefore ended up in a file left on the runner, or nowhere.

Changes

• Tests (datalab/tests/conftest.py): the faulthandler_enabled option is disabled in tests, so that pytest's faulthandler stays active and reports crash tracebacks in the test output. Checked locally: without this change, a crash in an application context prints nothing in the pytest output; with it, the traceback points to the crashing test line.
• PyQt5 workflow: the core dump of a crashed test process is kept, and its native backtrace (all threads) is printed by a step that runs on failure only.

Notes

• Three manual runs on this branch were green: the crash depends on garbage collection timing, which changes with the run metadata. It occurred on push runs on develop, so merging this is the best way to capture its traceback if it happens again.
• Diagnostics only: the crash itself is not fixed here.