Add a CSR batched matrix format, CUDA, HIP and DPCPP kernels

Add a CSR batched matrix format, CUDA, HIP and DPCPP kernels #1375