Keras MNIST dataset label mismatch and low accuracy, huge loss (for convolution)

Question

I have been trying to start learning DL and tried on Dense and Convolution. I got into trouble of low accuracy and high loss for convolution, while it is fine for dense, which I