A Novel Paradigm for Neural Computation: X-Net with Learnable Neurons and Adaptable Structure