First stable release of CNN from scratch for MNIST

🎉 First release of my Convolutional Neural Network (CNN) implementation from scratch using NumPy for the MNIST dataset.

✅ Features:

• Conv2D + MaxPooling + ReLU + Softmax architecture
• SGD with momentum optimizer
• Manual forward and backward passes
• Supports saving/loading model weights
• Includes inference on custom images

Metric	Value
Final Training Accuracy	~99%
Test Accuracy	~98%