Leveraging the hybrid CNN-BiLSTM and multi-head self-attention mechanisms for Aero-engine RUL estimation