kill 50ms sleep waits in client lib.rs tests

[intendednull]

Refs #271.

ten tokio::time::sleep(50ms) waits in crates/client/src/lib.rs get nuked. issue text say sleeps in production methods but file got reshuffled — they all in #[cfg(test)] mod tests block now (line 1255+). same bad pattern, same fix.

buckets per site

drop sleep, same StateActor, FIFO mailbox handle ordering (6 sites):

• send_message_and_read_back — send_message await apply_event await mutate(event_state); messages() read same actor
• switch_server_updates_event_state after send_message — same as above
• switch_server_updates_event_state after switch_server — switch_server await all four mutations (dag, event_state, server_registry, chat_meta) before return
• generate_invite_grants_send_permission_to_recipient — generate_invite end with apply_event await; test select same actor
• mutate_channel_mute_emits_event_and_flips_stats — sleep was BEFORE subscribe_events; broker only deliver to currently-subscribed recipients, so prior CreateChannel publish cant leak
• mutate_channel_mute_toggle_off_clears_set — three await mutations on same event_state actor

poll helper, derived actor recompute genuinely async (4 sites):

• presence_self_override_round_trip — mutate presence_meta, read view_handle.presence (derived from 3 sources)
• presence_reachable_peer_defaults_to_here — mutate chat_meta via peer_connected, read derived presence
• presence_queued_then_gone_after_threshold after peer_disconnected — same shape
• presence_queued_then_gone_after_threshold after tick burst — same shape

derived actor flow: Notify → spawn snapshot → do_send(UpdateCache) → cache replaced. get() return cached value, can be stale. fix = small await_view(probe) helper that loop get().await + yield_now() under 5s timeout. matches existing wait_for_message/wait_until pattern in tests/multi_peer_sync.rs.

why poll, not subscribe to Notify

subscribe AFTER mutation is racy (Notify may have already fired and dropped). fresh get cheap. tight loop bounded. simpler.

ask/watch buckets considered

issue mention ask(...) swap. only work when actor reply with () after applying mutation. ALL ten sites already going through state::mutate(...).await (which IS ask(Mutate) under the hood). so for same-actor reads, nothing to swap — sleep just defensive cargo cult. for derived-actor reads, ask(Get) return cached value not yet recomputed, so ask no help. polling necessary, NOT a fixed wait.

ran out of bucket=watch sites — no truly fire-and-forget sleep in this file.

gate

ran raw cargo (no just in sandbox):

• cargo fmt --check — clean
• cargo clippy --workspace --all-targets -- -D warnings — clean
• cargo test -p willow-client --lib — 303 pass before, 303 pass after
• presence tests run 3x in a row clean
• cargo check --target wasm32-unknown-unknown -p willow-client — clean
• cargo test --workspace --tests — all green

before/after grep

$ rg -c 'tokio::time::sleep' crates/client/src/lib.rs
# before: 10
# after:  0

https://claude.ai/code/session_01FHLLfYeh9P9FP7Y47wu9We

---

Generated by Claude Code