fix(daemon): remove pending broadcast on attach to prevent reattach double-render

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

crates/spaceshd/src/surface.rs

@@ -63,13 +63,11 @@ pub fn spawn_surface(
                         }
                         Some(SurfaceMsg::Attach { reply }) => { let _ = reply.send(bcast.subscribe()); }
                         Some(SurfaceMsg::AttachSnapshot { reply }) => {
-                            // Flush pending into the grid first so the snapshot is current,
-                            // but DO NOT broadcast here; subscribe before any further output.
-                            if !pending.is_empty() {
-                                grid.feed(&pending);
-                                let _ = bcast.send(std::mem::take(&mut pending));
-                                flush_deadline = None;
-                            }
+                            // Subscribe-then-snapshot is atomic within this actor turn (no await,
+                            // no flush can interleave). Any not-yet-flushed `pending` stays in
+                            // `pending` and is delivered to ALL subscribers — including this new
+                            // one — exactly once by the normal 6ms/16KiB flush path. It is NOT in
+                            // this snapshot. Broadcasting here would double-render on reattach.
                             let sub = bcast.subscribe();
                             let snap = snapshot_ansi(&grid);
                             let _ = reply.send((snap, sub));