applib/recognizer: drive the dead-zone fallback and make it geometric

[Mearman]

Draft — still validating on hardware.

Something else I noticed while experimenting with touch: on windows with a single registered widget (menu/scroll), any touch starting in the status-bar strip at the top of the screen did absolutely nothing — no scroll, no bridge buttons.

The router's dead-zone branch answers Tier-1 for the window's sole widget, but prv_resolve_widget_target only walks the active layer's parent chain, which by construction already failed (that's why the dead-zone branch ran). The latched target comes back NULL, the unified set is failed at the latch, and tier exclusion has already failed the bridge set — so both sets are dead and the touch drives nothing.

Two changes:

• prv_resolve_widget_target now resolves the dead-zone fallback too, so the Tier-1 route actually has a driver, and the unified set's touch filter is gesture-stable (fresh resolve on the Touchdown, the latched route afterwards) so dragging out of the strip doesn't drop the widget mid-gesture.
• The fallback is geometric on both sides: the sole widget captures the touch only when its layer frame contains the Touchdown point. The dead zone exists for content that continues beneath the status bar; a widget positioned away from the strip has nothing under the touch and must not capture it — which test_scroll_layer_touch__dispatch_no_trigger_leaves_manager_idle already documented in intent, and which pins the behaviour here.

New tests: dead-zone Touchdown latches and drives an ops-bearing sole widget through a full pan, and a sole widget away from the strip drops. Full unit suite green.