removed the voice agent

.env.example

@@ -55,21 +55,3 @@ RATE_LIMIT_MAX=40
 # --- Artifacts ---
 ARTIFACT_DIR=./storage/artifacts
 ARTIFACT_TTL_HOURS=72
-
-# --- Voice (speech-to-speech, CPU, in-process) ---
-# Models are ONNX via Transformers.js — no GPU, no Python, no second service.
-#   STT  onnx-community/whisper-tiny.en  English only; half the RAM of base
-#   TTS  Xenova/mms-tts-eng              from the Hub
-# Tamil is built and tested (assets/tts/mms-tts-tam, exported locally — no
-# public ONNX exists). Enable it with VOICE_LANGUAGES=en,ta and a multilingual
-# STT_MODEL, but budget ~200 MB more resident for the extra voice.
-VOICE_LANGUAGES=en
-SPEECH_WARMUP=false
-STT_MODEL=onnx-community/whisper-tiny.en
-# Below this confidence, the user's preferred language beats the detector.
-DETECT_CONFIDENCE=0.6
-
-# Endpointing
-VAD_SILENCE_MS=700
-VAD_MIN_SPEECH_MS=250
-VAD_PREFIX_MS=300