Add dynamic r2s/s2r copy function.

[ChengXiang Qi]

corresponds to these process:

1. declare (??) and instantiate the copy plan

TiledCUDA/src/kernels/cute_gemm.cu

Lines 50 to 52 in 8205e7c

 auto rA = make_s2rA(sA_ptr, tid, typename KeTraits::SmemLayoutA{}, mma); 
 auto rB = make_s2rB(sB_ptr, tid, typename KeTraits::SmemLayoutB{}, mma); 
 auto acc = get_acc<kTM, kTN>(mma); 

2. execute copy in time

TiledCUDA/src/kernels/cute_gemm.cu

Lines 64 to 66 in 8205e7c

 for (int i = 0; i < rA.get_iters(); ++i) { 
     rA.copy(i);  // load A register tile from shared memory 
     rB.copy(i);  // load B register tile from shared memory