Synthesizing Specialized Sparse Tensor Accelerators For Reconfigurable Hardware Using High-Level Functional Abstractions