Misplaced alignas() in cuda/memcpy_async/memcpy_async_tx.pass.cpp

https://github.com/NVIDIA/cccl/blob/4e6f24a6af6b661ed561f1e598df109f03973cdf/libcudacxx/test/libcudacxx/cuda/memcpy_async/memcpy_async_tx.pass.cpp#L42

`alignas(16)` should be placed before `__shared__`

Both nvcc and clang complain about misplaced `alignas(16)`, though for nvcc it's a warning, but it's an error for clang: https://godbolt.org/z/YqfGTznYf