feat(ai): server-backed conversation history (objectui parity)

[os-zhuang]

What

Brings the mobile AI assistant to web (objectui) parity: when the server exposes /api/v1/ai/conversations, chats are saved server-side — multi-conversation, cross-device, listable — instead of only an on-device cache.

• lib/ai-conversations.ts — typed client for the conversation API (list/create/get/delete + addMessage) and a cached conversationsAvailable() probe so servers without it (published service-ai 8.0.1) degrade cleanly.
• lib/ai-chat.ts — thread conversationId through stream/sendAiChat request bodies.
• stores/ai-chat-store — two-tier persistence:
  • Server mode: create a conversation on first send, carry conversationId into /ai/chat, explicitly persist user+assistant via /:id/messages, list/load/delete, resume the cached active id on open.
  • Local mode (no API): the existing MMKV restart-survival cache (PR feat(ai): persist the conversation to disk (survives app restart) #97).
  • init() probes once and picks the mode.
• app/ai.tsx — a Conversations drawer (new chat / switch / delete) + history actions in server mode; calls init() on mount.

Verification

• End-to-end against the framework dev server (which ships the conversation API): create → chat(conversationId) (streamed a real LLM reply) → messages persist → GET /:id returns the thread → GET /conversations lists it. Confirmed the exact store sequence works server-side.
• The published 8.0.1 server lacks the API → the demo falls back to the local cache (graceful, no breakage).
• 48 AI tests (conversation client incl. availability probe; store server + local modes — create-on-send, persist, load, delete, new; drawer UI load/delete).
• npm run lint 0 errors; jest 1249/1249 (152 suites).

Notes

• The server-side conversation API is currently only in the unreleased framework source (published @objectstack/service-ai 8.0.1 has none) — so server-backed mode activates automatically once a server with /conversations is connected; until then mobile uses the local cache.
• Persistence uses the explicit /:id/messages endpoint (reliable across server builds) rather than relying on chat-endpoint auto-persist.

🤖 Generated with Claude Code