Fix: Multi-window UI tests can produce inconsistent results

After creating a new scene in a separate window when performing UI
tests, the respective view layer for the window may not be updated
immediately in the event loop.

Previously, this was mitigated with a single `yield` statement that
would delay processing by a single tick. To fix this issue, this commit
adds the capability to yield for a specific `timedelta` and waits this
amount of time for the two affected tests.

Pull Request: https://projects.blender.org/blender/blender/pulls/136012

tests/python/ui_simulate/test_undo.py

@@ -5,6 +5,7 @@
 """
 This file does not run anything, it's methods are accessed for tests by: ``run.py``.
 """
+import datetime
 
 # FIXME: Since 2.8 or so, there is a problem with simulated events
 # where a popup needs the main-loop to cycle once before new events
@@ -653,7 +654,10 @@ def view3d_multi_mode_multi_window():
     yield from _call_menu(e_b, "New Scene")
     yield e_b.ret()
     if _MENU_CONFIRM_HACK:
-        yield
+        # We wait for a brief period of time after confirming to ensure that each main window has a different view layer
+        yield datetime.timedelta(seconds=1 / 60)
+
+    t.assertNotEqual(window_a.view_layer, window_b.view_layer, "Windows should have different view layers")
 
     for e in (e_a, e_b):
         pos_v3d = _cursor_position_from_spacetype(e.window, 'VIEW_3D')
@@ -809,7 +813,10 @@ def view3d_edit_mode_multi_window():
     yield from _call_menu(e_b, "New Scene")
     yield e_b.ret()
     if _MENU_CONFIRM_HACK:
-        yield
+        # We wait for a brief period of time after confirming to ensure that each main window has a different view layer
+        yield datetime.timedelta(seconds=1 / 60)
+
+    t.assertNotEqual(window_a.view_layer, window_b.view_layer, "Windows should have different view layers")
 
     for e in (e_a, e_b):
         pos_v3d = _cursor_position_from_spacetype(e.window, 'VIEW_3D')