fix(ai-agent): team model selector fallback from registry cache

[zhuqingyv]

Summary

• Fix: team mode model selector UI disappears because GET /api/conversations/:id/model returns available_models: [] and current_model_label: null
• Root cause: ACP session handshake (session/new) only triggers on first send_message, but the frontend queries the model endpoint immediately after agent warmup
• Fix: when session live data is empty, fall back to the AgentRegistry's cached handshake data (maintained by CatalogForwarder)

Design rationale

Respects the existing three-layer state model (Desired/Observed/Advertised) — no schema changes, no DB migration, no preload semantic changes. Uses the existing CatalogForwarder → Registry data flow as the pre-handshake fallback source. Once the CLI completes its handshake, live data automatically supersedes the fallback.

Changes

• crates/aionui-api-types/src/acp.rs: Add Deserialize to ModelInfoEntry (enables parsing from registry JSON)
• crates/aionui-ai-agent/src/service.rs: Add registry fallback logic in get_model + private model_info_from_registry helper

Test plan

• cargo clippy --workspace — 0 warnings
• cargo test --workspace — all pass
• cargo fmt --all -- --check — pass
• Manual: create team → member warmup → immediately call GET model endpoint → should return non-empty available_models

🤖 Generated with Claude Code