io_uring: use member filling in io_fill_cqe_req()

io_uring: use member filling in io_fill_cqe_req()

At least on AMD architectures, manually filling the few fields we need
to copy ends up stalling a lot less than calling memcpy() to do the
same thing. Since it's just 3 fields here, do it manually.

Don't both with the big cqe, just do this for the normal CQE filling.

Signed-off-by: Jens Axboe <axboe@kernel.dk>