Fix AttributeError on iconlist when GalleryTab parent dialog closes (bug 13326)

[eduralph]

Root cause

GalleryTab.clean_up() calls super().clean_up(), which removes
self.iconlist via track_ref_for_deletion("iconlist"), but never
disconnects the 'selection-changed' signal handler on the iconlist
widget. When the parent dialog tears down on Cancel — the Forms
addon's GalleryTab-in-a-dialog path exercises this reliably — GTK
dispatches a final selection-changed to the still-live widget; the
handler calls self.get_selected() → self.iconlist.get_selected_items()
and raises AttributeError: 'GalleryTab' object has no attribute 'iconlist'.

Fix

Capture the 'selection-changed' handler id in build_interface,
then disconnect it at the head of clean_up() — before
super().clean_up() removes the attribute. Guard the disconnect
with GObject.signal_handler_is_connected so the already-disposed
case (the reason the original disconnect was removed in 37395da)
stays silent.

Verified against

• gramps/gui/editors/displaytabs/gallerytab.py:232,659-660 — the
  unguarded connect and the no-op clean_up that the reported
  trace fingers.
• gramps/gui/editors/displaytabs/buttontab.py:319 — the call site
  in the inherited _selection_changed callback that reaches into
  self.get_selected().
• The 2012 disconnect (svn r20849 / 51a53ccebd) — same defect
  class, same fix shape; removed in 37395da262 to silence a GTK
  warning on the normal close path. The new
  signal_handler_is_connected guard keeps that path silent while
  restoring teardown-race safety.

Test

gramps/gui/editors/displaytabs/test/gallerytab_test.py covers both
the bug and the warning the original disconnect was removed to
silence:

• test_clean_up_disconnects_selection_changed constructs a real
  GalleryTab, records the 'selection-changed' signal id, and
  asserts via GObject.signal_handler_find that no handler remains
  after clean_up(). Pre-fix 42 != 0; post-fix passes.
• test_clean_up_after_iconlist_destroyed_emits_no_warning destroys
  the iconlist widget before clean_up() — the normal close path
  where commit 37395da was seeing the GTK critical — and asserts
  via warnings.catch_warnings that no instance '...' has no handler with id '...' warning is recorded. Mutating the fix to
  drop the signal_handler_is_connected guard surfaces exactly that
  warning, so future removals of the guard regress here.

Both need a display (Gtk.IconView), so run under xvfb-run:

xvfb-run -a python3 -m unittest \
    gramps.gui.editors.displaytabs.test.gallerytab_test -v

Fixes #13326.

[romjeromealt]

Is it also related to bug report #12289?

[eduralph]

Thanks for the pointer to 0012289 — it's relevant to understand the story behind this one.

The key distinction is severity and path. 12289 was a cosmetic GObject warning (instance ... has no handler with id ...) on the normal Person Editor close. 13326 is different in kind: an unhandled AttributeError crash on a specific sequence — add an image to the gallery, then Cancel the dialog — where a selection-changed callback fires after iconlist has been torn down.

This resolves the crash without re-touching the part of the cleanup behavior that the earlier reports were wary about. Happy to adjust the approach if there's a side effect I've missed — the guard is specifically there to respect the original concern,

[romjeromealt]

Do not need to worry too much, I often make relations, this does not mean something should be improved. I am fine with this PR. As far as I remember the root of 12289 was pointed during a tracking down session around memory leaks. So, even cosmetic GalleryTab or some gramplets, generated some minor memory issues. Now, with the help of AI, we can just provide something more "professional[1]", quick and maybe more efficient. That's all.
[1] my workflow was to run Gramps 5.1.3 with an OS using less than 512Mo of ram... and I closed 12289 as "wont't fix" because today, such config would not run Gramps 6 and later.

[Nick-Hall]

Added 2 files to POTFILES.skip.

We probably should run our CI unit test suite using Xvfb, but that can be done in a separate PR.