FastCache: Fast Caching for Diffusion Transformer Through Learnable Linear Approximation | Read Paper on Bytez