feat(app): integrate realtime agent into the agent service in `agentscope.app`

**Background**
`agentscope.app` currently has no support for realtime voice interaction. As the realtime agent implementation matures, a corresponding frontend and backend is needed to expose these capabilities end-to-end.

**Changes**
- Backend: add a WebSocket-based audio streaming endpoint with support for distributed deployment and user-level session isolation
- Frontend: build a minimal voice interaction UI covering mic input, audio playback, and session control
- Wire up to both speech-to-speech and cascaded `RealtimeModel` backends