Unlocking Deep Learning: A BP-Free Approach for Parallel Block-Wise Training of Neural Networks