Post task updates to the loop instead of waking it

`WakeTaskQueue` becomes `TaskQueue`, which carries the update itself.
Delivery and waking are then one act: the winit host sends it through the
`EventLoopProxy` as a `DefaultEvent::Update`, so there is no channel
beside the loop and nothing has to claim a redraw is needed in order to be
looked at. `Window::request_redraw` is gone from this path; `event`
applies the update and then asks the tree whether anything became dirty,
which is the same question `window_event` already ended with -- now
`schedule_redraw`, called from both.

The loop's message type is `DefaultEvent<State>`, so `Proxy` becomes a
wrapper that takes the application's own `Event` and requires it to be
`Send`, since it now crosses to the task thread by that route.

The harness supplies a channel-backed queue, which is what lets a test
hold updates until it asks for them.

Tests split by subject -- layout, pointer, scroll, tasks -- with the
region helper in `tests/common`.

examples/task.rs

@@ -11,11 +11,7 @@ struct State {
 }
 
 impl DefaultAppState for State {
-    fn new(
-        mut ui_state: DefaultUiState,
-        rsc: &mut DefaultRsc<Self>,
-        _: Proxy<Self::Event>,
-    ) -> Self {
+    fn new(mut ui_state: DefaultUiState, rsc: &mut DefaultRsc<Self>, _: Proxy<Self>) -> Self {
         let rect = rect(Color::RED).add(rsc);
         rect.task_on(CursorSense::click(), async move |mut ctx| {
             tokio::time::sleep(Duration::from_secs(1)).await;