docs(gpui): correct async and test context docs

[kirillDevPro]

What & why

crates/moon-gpui/docs/contexts.md described async and test contexts in a way the code does not.

• It said calling to_async makes entity calls fallible because the context can outlive the app. AsyncApp holds a weak app reference. Entity::update on that context panics if the app is already gone (app().expect(...)) and does not return Result. AsyncApp::update_window and AsyncWindowContext::update return Result when the app was dropped, the app is quitting, or the window is gone.
• It said TestAppContext panics on a missing app or window. The context owns the app through Rc. TestAppContext::update_window returns Result when the window is gone. VisualTestContext::update unwraps that result and panics.
• It said WindowHandle::update yields a Window. The method takes an AppContext and a closure FnOnce(&mut V, &mut Window, &mut Context<V>) -> R and returns Result.

How to verify

Markdown only. Compared the new sentences with App::to_async, Window::to_async, Context::spawn / spawn_in, AsyncApp::update_entity / update_window, AsyncWindowContext::update, Entity::update, WeakEntity::update, TestAppContext, VisualTestContext::update, and WindowHandle::update.

A clean-context review disputed three claims. Those parts were dropped or narrowed before push, and the diff was not sent back for a second reading:

• Context<T> does not lack to_async. It dereferences to App, so cx.to_async() is App::to_async.
• Not every app-using AsyncApp method panics. The panic claim is limited to Entity::update. Window updates return Result, including when the app was released.
• TestAppContext has no clock helpers of its own. advance_clock lives on BackgroundExecutor. The clock sentence was removed.