Vectorized 1-NN and Neural Network Forward Pass

Compute squared Euclidean distances for training and query arrays without loops in NumPy, then map argmin logic to Wx+b layers and softmax for OpenAI.