feat(voice): wire composed voice into gateway and prewarm TTS

[yyy-router]

Part of #312

把组合式语音接入网关并做 TTS 预热（依赖 #319 的组合式编排器）：

• composition.py：build_composed_voice_agent 组合根，装配 ASR + LLM + TTS + Agent
• composed_audio.py：ComposedAgentAudioSink 网关适配器（consume / interrupt / close_session）
• ports.py：新增 AudioSessionLifecycle 会话生命周期契约
• voice_stream.py：新流触发 interrupt、断连触发 close_session
• main.py：TIMEFLOW_VOICE_AGENT_MODE=2 分派到组合式 agent，移除原 mode-2 的 RuntimeError 占位
• pipeline.py + qwen_audio_tts.py：speech_expected 提前建立 TTS 连接，WebSocket 握手与分词并行
• settings.py：VAD 阈值 0.0 → 0.6，抑制背景噪声误触发

[codecov]

Codecov Report

❌ Patch coverage is 95.04950% with 5 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...ckend/src/timeflow/intelligence/speech/pipeline.py	83.33%	1 Missing and 2 partials ⚠️
backend/src/timeflow/composition.py	96.87%	1 Missing ⚠️
...flow/infrastructure/external/tts/qwen_audio_tts.py	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[fennoai]

Reviewed the fixed 6a896276a1950546c0f2e1b6b26f82cc73ec4085...945e2e408a37d5fe3e9af2ea3106ab8bc2a1b52f diff across composition-root wiring, gateway lifecycle hooks, speech/TTS prewarming, settings, and associated tests. I found no actionable correctness, reliability, security, or compatibility regressions meeting the review threshold.

Validation: python3 -m compileall -q backend/src backend/tests and git diff --check passed. The focused pytest suite could not run because pytest and uv are not installed in the workspace.

[gac0812]

ok