Add warmup process for VAD model initialization

Initialize the VAD model with a warmup process using zeros to ensure proper functioning before processing actual audio data.

webrtc_vad.py

@@ -56,6 +56,9 @@ class RealtimeVAD:
         self.sum_positive_s = 0.0
         self.silence_start_s: float | None = None
 
+        # Warmup
+        self.vad_model.process(np.zeros(hop_size, dtype=np.int16))
+
     def process(self, audio_data: np.ndarray):
         if audio_data.ndim == 2:
             # FastRTC style [channels, samples]