[2406.00699] Towards General Robustness Verification of MaxPool-based Convolutional Neural Networks via Tightening Linear Approximation