v0.4.0

Remote Pi v0.4.0 — voice, images, and a big app/UX round.

Tag tracks the pi-extension version (0.4.0). 37 commits since v0.2.1.

📷 Images & 🎙️ Voice

• Image attachments end-to-end: protocol + multimodal ingest (pi-extension) and composer attach (app).
• Voice input via native speech-to-text in the composer.
• Fix — image send unblocked: relay envelope size limit is now configurable (RELAY_MAX_CT_MIB, default 4 MiB). The previous fixed 1 MiB cap silently dropped larger images, leaving them stuck on "sending…".
• App durable send timeout: a message that never gets echoed is removed silently after 20s, and the timeout re-arms on session load/restart.
• Images survive session_sync replay (history mapper).

📱 App layout & UX (plan-32)

• Working room, chat polish, session-switch, detail-pane, tool-card, settings URL.
• Working/thinking indicators decoupled from the stream; thinking cursor restored at turn start.
• Device classification by shortestSide — a phone in landscape no longer flips into tablet mode.
• Compaction feedback (working bracket + result broadcast + history); forced English compaction summary.

🗄️ Architecture

• Local DB as the single source of truth (reactive SSOT) — plan-31.
• working field in room_meta, published on turn start/end (relay + pi-extension).
• Transport: liveness watchdog on RelayClient (force-close after silence).
• Quick Actions from the mobile app (plan-28); persist model_set to project settings (survives Pi/daemon restart).
• Slash command handling + command mirror for SDK built-ins.

🔧 Chores & platform

• Version bumps: pi-extension 0.4.0, relay 0.2.2, app 1.1.0+5.
• iOS deployment target 18.0; ML Kit / CameraX 16 KB page-size compliance; App Store / Google Play store-listing metadata.

Full Changelog: v0.2.1...v0.4.0