"""Step 1 (batch): Generate synthetic audio samples from a CSV manifest. Reads rows from a CSV file and generates one WAV per row, loading IndexTTS2 only once for efficiency.