fix: Stack takes part in focus traversal; ListBox clips wide rows

[tannevaled]

Two small fixes found while building the claimward Linux and Windows apps, one commit each.

Stack (#480)

Root cause. Stack keeps its pages in a map[string]Widget and exposed neither Children() nor focusableChildren(), so every tree walk stopped at it:

• Keys: a click focused an Entry inside the page (the page's own container ran focusClick), but the enclosing container's focus list never contained it, and it is that container which routes every key, so typed keys went to the wrong list and were dropped.
• Popover: PopoverHost finds open popovers through Children(), so a DropDown on a Stack page opened a popover nobody drew on top or routed clicks to.

Fix. Stack yields its visible page to both walks and its OnEvent now runs routeFocusKey and click-to-focus before forwarding, as Container/AlignBox do. Hidden pages are deliberately not yielded: unlike an inactive CardLayout card (collapsed to an empty rect, which gatherFocusables skips), a hidden Stack page keeps the bounds it last had, which are the Stack's own, so a bounds check cannot exclude it. Yielding it would make Tab reach invisible widgets, put them in the accessibility tree where the visible page is, and let an open popover on a hidden page take clicks.

Guard. TestEveryContainerExposesItsChildren only recognised Widget and []Widget fields, which is how Stack got through. It now also treats map[K]Widget as a holder (no other widget in the package holds one; ViewController.refs is not a widget).

Tests (stack_focus_test.go): typed keys reach an Entry on the visible page under a root VBox; Tab walks outside field → visible page and never into a hidden page with stale bounds, then into the new page after SetVisible; PopoverOwners reports the open DropDown on the visible page and nothing once it is hidden; a Stack used as the root routes focus itself; Children/focusableChildren yield only the visible page. Plus a stack case in the existing Children tests.

ListBox (#481)

Root cause. ListBox.Draw (flat and sectioned) pushed its clip only if overflow, a clip meant to stop a partial trailing row bleeding past the bottom. A list that fits vertically was not clipped at all, so a row wider than the list drew past its right edge (a host ItemRenderer too).

Fix. The clip to the content rect is pushed whenever the painter is a painter.Clipper, as Entry and SearchEntry already do. It is the same rect as before, so the overflowing case renders the same.

Test (list_clip_test.go): pixel render on a PixelPainter wider than the list; no pixel right of the list's edge may be touched, for flat-fits, flat-overflows (control, passed before), sectioned-fits, and an ItemRenderer that paints 200 px past its rect.

Verification

• Before the fix (original stack.go / list.go, new tests): TestStackEntryReceivesTypedKeys, TestStackTakesPartInTabTraversal, TestStackExposesItsVisiblePagePopover, TestStackAsRootRoutesFocus, TestEveryContainerExposesItsChildren ("Stack holds Pages") and the three non-overflow cases of TestListBoxClipsRowsWiderThanTheList fail; the overflow case passes, as expected.
• Sabotage, each passing go vet: Children yielding every page fails the hidden-page assertions (Tab, popover, Children); dropping routeFocusKey/focusClick from Stack.OnEvent fails TestStackAsRootRoutesFocus.
• Locally: gofmt clean, go vet ./..., go test ./... with every function at 100.0% coverage, -race on the touched tests.

README does not describe either behaviour, so it is unchanged; the doc comments of Stack, ListBox.Draw and focusEnumerator are updated.

Semver

Minor. (*Stack).Children() is new exported API (it makes *Stack satisfy the child-walk convention). The rest is a behaviour fix.

Known, not changed here: focus held by a widget on a page that SetVisible hides is not cleared, the same as an inactive CardLayout card today.

Fixes #480
Fixes #481

🤖 Generated with Claude Code