Support MLP, KAN, and KAN derivatives

KAN paper: Kolmogorov-Arnold Networks

KANs:

• KANs
• Efficient-KAN
• FourierKAN
• Two-layer MLP (Toy Version)

python examples/mnist.py --model MLP
python examples/mnist.py --model KAN
python examples/mnist.py --model MNISTFourierKAN

• For KAN, a large initial learning rate may be more effective. (You can try lr = 1e-2)
• In my experiments, KAN does converge faster than MLP.