Meeting recording capsule won't stay in a new corner after drag (Linux/Windows) #1767

What I see

The black recording pill (Meeting — 8:55 AM / elapsed time, screenshotted on Linux) blocks something on screen. I drag it, it follows the pointer, but on release it jumps back to the corner it started in instead of staying where I dropped it.

Why

Capsule::drag (src/rune_ui/meeting_capsules.rs) snaps a released drag with primary_screen_size().map_or(self.corner, ...). On macOS that reads NSScreen; on every other OS it is hardcoded to None, so the fallback is always the starting corner:

/// Linux and Windows leave a dragged capsule in its corner.
#[cfg(not(target_os = "macos"))]
fn primary_screen_size() -> Option<(f32, f32)> {
    None
}

So on Linux/Windows the capsule can never move corners — release always snaps back.

Diagnosis for the fix

• Linux: the X11 root window size (x11rb::connect + setup().roots, already a dependency) is the screen Rune places the capsule on. XWayland answers for its own root window; on pure Wayland there is no X server, so None (stays in corner) is correct there.
• Windows: GetSystemMetrics(SM_CXSCREEN/SM_CYSCREEN) (windows-sys, already a dependency).
• The pure snap math (nearest_corner) is already OS-independent and covered by a_released_capsule_snaps_to_the_corner_nearest_its_centre.

Suggested patch

I have this working on a branch with cargo check clean, 26 meeting_capsules tests + 46 labs tests passing, fmt + code-shape clean. Diff sketch:

#[cfg(target_os = "linux")]
fn primary_screen_size() -> Option<(f32, f32)> {
    use x11rb::connection::Connection as _;
    let (connection, screen) = x11rb::connect(None).ok()?;
    let size = connection.setup().roots.get(screen)?;
    (size.width_in_pixels > 0 && size.height_in_pixels > 0).then(|| (
        f32::from(size.width_in_pixels),
        f32::from(size.height_in_pixels),
    ))
}

#[cfg(target_os = "windows")]
fn primary_screen_size() -> Option<(f32, f32)> {
    let width = unsafe { GetSystemMetrics(SM_CXSCREEN) };
    let height = unsafe { GetSystemMetrics(SM_CYSCREEN) };
    (width > 0 && height > 0).then_some((width as f32, height as f32))
}

Happy to post the full diff (4 files: the fix + a stale-comment update in rune_ui/labs.rs + docs/features/{meetings,labs}.md updates) wherever is most useful — per CONTRIBUTING I am not opening a PR.

Environment: Linux (X11), Hex at dev 1982f53.