Serialize graph-batch on the walker thread, not the UI thread

[zangjiucheng]

The app's hottest IPC stream serializes on the main thread. commands.rs:293 calls event_util::emit_on_main, whose closure at event_util.rs:31-36 runs handle.emit on the main thread; tauri 2.11.5's emit wraps the payload as EmitPayload::Serialize and serializes downstream in EmitArgs::new.

Honest scale

typical batch (1000 rows / 5k gaps)	0.22 ms
worst case (MAX_GAP_SEGMENTS_PER_BATCH = 200_000, commands.rs:113)	2.7 ms

against an 8 ms throttle floor.

So this is worth doing because the walker thread is idle anyway and the change is small — not because it is a large main-thread stall. Sizing it honestly matters more than shipping it.

Ride-along

Fix the stale doc at event_util.rs:11,21-22, which still claims graph-batch uses an ipc::Channel. It does not — that approach was abandoned because it broke multi-window and second-instance handling.

Sequencing

Land after the wide-branching perf test, so it goes in under a regression guard.

Land after #84 so it goes in under a regression guard.

---

_{Found during v1.3.0 research: parallel codebase surveys, then each finding independently verified against the code before filing. Line numbers are from dev at the time of writing.}