fix(backend): drop the prompt cache before the allocator-empty; note clear_cache is thread-agnostic

Review follow-ups: clear_voice_prompt_memory_cache() now runs before
backend.unload_model() at all three call sites so the device-resident
prompt tensors are already released when empty_device_cache() /
empty_mlx_cache() runs, instead of going back into the caching allocator
afterwards. empty_mlx_cache's docstring states why it may run off the
MLX worker thread.

backend/services/tts.py

@@ -24,8 +24,8 @@ def get_tts_model() -> TTSBackend:
 def unload_tts_model():
     """Unload TTS model to free memory."""
     backend = get_tts_backend()
-    backend.unload_model()
     clear_voice_prompt_memory_cache()
+    backend.unload_model()
 
 
 def audio_to_wav_bytes(audio: np.ndarray, sample_rate: int) -> bytes: