docs(audits): stdlib sync channel overhead in transfer hot path

[oferchen]

Summary

• Survey of remaining std::sync::mpsc and Arc<Mutex<channel>> sites on the receiver-side transfer hot path after PR Explicitly disable Windows x86/aarch64 cross targets #1744 migrated work_queue.rs to crossbeam_channel::bounded.
• Per-channel-op cost table comparing std, crossbeam, crossbeam_queue::ArrayQueue, and parking_lot::Mutex<VecDeque<T>> backends.
• Criterion bench plan (crates/engine/benches/channel_overhead.rs) sweeping 1/4/16/64 producers across each backend, with end-to-end checks on real receiver workloads.
• Migration recommendations: highest-confidence target is the Arc<Mutex<mpsc::Receiver>> shared queue at crates/signature/src/async_gen.rs:218 (site B). Two further bounded-channel migrations conditional on bench data. No parking_lot migration recommended.

Test plan

• Doc-only change; rendering verified locally.
• No code or test impact; CI fmt+clippy must still pass on existing tree.