Fix: audio_output Hang + Stimme auf Seraphina

- speak() hing weil pygame.time.wait() die asyncio Event-Loop blockierte.
  Fix: time.sleep(0.05) statt pygame.time.wait() im Playback-Loop.
- asyncio.new_event_loop() statt asyncio.run() — thread-sicher, kein
  "event loop already running" in Worker-Threads.
- _speak_lock verhindert parallele TTS-Aufrufe.
- Stimme: de-DE-SeraphinaMultilingualNeural als Default (modern, weiblich,
  multilingual neural). Dropdown in Settings mit allen DE/EN Optionen.
- speak()-Signatur: voice + rate als Parameter (statt config-Import).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

client/hermes-voice/config_manager.py

@@ -9,7 +9,7 @@ DEFAULTS = {
     "chat_model": "Hermes-4-14B",
     "vision_model": "Qwen3-VL-2B-Instruct",
     "hotkey": "ctrl_r",
-    "tts_voice": "de-DE-KillianNeural",
+    "tts_voice": "de-DE-SeraphinaMultilingualNeural",
     "tts_rate": "+0%",
     "whisper_model": "small",
     "language": "de",