[PierreRaybaut]

Main windows created by tests were destroyed at an arbitrary point of the next test, possibly amid Qt operations. This is a plausible cause of the intermittent segmentation fault seen on develop after #402 (in side_by_side_app_test.py, right when the previous main window became garbage), although it is not proven: see #405 for the diagnostics.

What kept a closed main window alive

Reference	Released
SigimaX singleton instance (SGMXMainWindow.get_instance)	when the next main window was created
Web API controller (datalab.webapi), shared by all main windows	when the next main window was created
Closed internal console (WA_DeleteOnClose), whose namespace references the main window	by the next event loop, i.e. during the next test

The main window being part of reference cycles, it was then only destroyed when the garbage collector happened to run.

Changes

• Web API: when its main window is closed, WebApiActions.cleanup disconnects its actions from the shared controller (they could otherwise be notified by a server started from another window) and the controller releases the window (release_main_window). Stopping the server also releases the workspace adapter set for the routes, which references the main window.
• Tests (conftest.py): after each test, pending events and deferred deletions are processed, then the garbage collector is run.
• Requires Release the closed main window as singleton instance SigimaX#11 (same branch name, used by CI): a main window closed for good is no longer kept as singleton instance.

Validation

• With all changes, no main window of a test is alive at the start of the next one (it was before, even after gc.collect()).
• Full test suite passes locally (1028 passed, 3 skipped). Running deferred deletions and the garbage collector after each test adds about 7% to the duration (290 s → 310 s on features/image).