Tinycast translation result buttons

Midscene CI evidence for abue-ammar/tinycast#1068. Each report uses startObserving() and window-level frame sampling while a button is held down.

Before

Hover and pressed captures are byte-identical for all three buttons.

After

All three buttons have a distinct pressed state; actions still execute.

The actions execute in both builds. Each report records aiAct, five continuously observed frames, and five panel-only held-mouse frames for a direct visual comparison.