Flake watch: fullscreen-holds-the-view failed once under heavy load

Observed during #90's gates: scripts/uitest.sh run 1 reported 'fullscreen holds the view' FAILING (view reset on f), run 2 clean 18/18 - on a branch touching neither camera nor fullscreen, while the box ran three concurrent builds at ~7 fps presented. Two readings: (a) load-induced harness timing (the f press or capture raced the redraw), or (b) the #77 fix has a real timing window where the widget-state move does not protect the view under extreme frame starvation. Per the negative-results doctrine this stays open until an idle-box rerun either reproduces it (then it is a real bug and gets the repro-first treatment) or a load-test deliberately reproduces it and attributes it to the harness. Do not close on 'could not reproduce' without a positive-capable instrument.