docs: Explain recovering focus when the focused element stops rendering

[AncientPixel]

Description

The Focus guide does not say what happens when a view stops rendering the element that holds Focus, for example when a table is replaced by an empty state. In an app this shows up as "keyboard shortcuts stop working after a view is hidden".

What GPUI does (gpui-pre 0.3.8, the version pinned on main):

• The window keeps the old FocusId. Window::focus_node_id_in_rendered_frame (src/window.rs:6325) looks it up in the rendered dispatch tree and, when the element was not rendered, falls back to dispatch_tree.root_node_id(). Key and action dispatch then walks the path to the root node only, so bindings under the view's key_context and its on_action handlers are skipped. Only handlers on the root node and global cx.on_action handlers still run.
• GPUI already has the recovery hooks. After a frame is drawn, if the previous focus path was non-empty and the new one is empty, the window runs the focus_lost_listeners (src/window.rs:3406). Context::on_focus_lost(window, listener) (src/app/context.rs:616) registers one. While those listeners run, Window::focus_lost_restore_target(cx) (src/window.rs:2337) returns the closest ancestor of the lost element that is still rendered and focusable.

The change:

• website/docs/focus.md: a short section, "When the focused element stops rendering", that explains the fallback and shows a root view restoring Focus with on_focus_lost and focus_lost_restore_target. A new row in the "Verify and debug" table for the symptom.
• website/zh-CN/docs/focus.md: the same section and table row in Chinese.
• skills/gpui-kit/references/gpui/focus-handle.md: the same pattern as item 4 under "Common Patterns". This file is source, not one of the vendored copies of website/docs/ that AGENTS.md lists.

This is a documentation-only change. It was written with AI assistance (Claude Code).

How to Test

1. Read the new section in website/docs/focus.md and website/zh-CN/docs/focus.md, and the new table row in each.
2. The example compiles. I put it in a scratch crate that depends on this branch's crates/kit by path (so gpui-pre 0.3.8), with a Render impl and main around it. cargo check passed with no errors and no warnings. The crate's src/main.rs:

src/main.rs

use gpui_kit::*;

struct Library {
    focus_handle: FocusHandle,
    _focus_lost: Subscription,
}

impl Library {
    fn new(window: &mut Window, cx: &mut Context<Self>) -> Self {
        let focus_lost = cx.on_focus_lost(window, |this, window, cx| {
            let target = window
                .focus_lost_restore_target(cx)
                .unwrap_or_else(|| this.focus_handle.clone());
            target.focus(window, cx);
        });

        Self {
            focus_handle: cx.focus_handle(),
            _focus_lost: focus_lost,
        }
    }
}

impl Render for Library {
    fn render(&mut self, _: &mut Window, _: &mut Context<Self>) -> impl IntoElement {
        div()
            .key_context("Library")
            .track_focus(&self.focus_handle)
            .child("No cards yet")
    }
}

fn main() {
    gpui_kit::application().run(|cx| {
        gpui_kit::init(cx);
        gpui_kit::open_window(WindowOptions::default(), cx, |window, cx| {
            cx.new(|cx| Library::new(window, cx))
        })
        .expect("Failed to open window");
    });
}

3. The website docs tests (bun run test:docs) and typos were not run: neither bun nor typos is installed on my machine. The change adds no links and no version strings.

Checklist

• I have read the CONTRIBUTING document and followed the guidelines.
• Reviewed the changes in this PR and confirmed AI generated code (If any) is accurate.
• Passed cargo run for story tests related to the changes. Not applicable: documentation only.
• Tested macOS, Windows and Linux platforms performance (if the change is platform-specific) Not applicable: documentation only.

🤖 Generated with Claude Code

https://claude.ai/code/session_01PRpJSFk3zTwdgAJHehdZFm

[AncientPixel]

The failing GPUI Fast (windows-latest) check isn't from this change. It fails at link time with LNK1123: failure during conversion to COFF, the same way it fails on main at c0bebdc, the commit this branch is based on. #3380 looks like the fix. Once it lands I'll rebase onto main so CI runs again.