__io_queue_sqe() tries to issue as much requests of a link as it can,
and uses io_put_req_find_next() to extract a next one, targeting inline
completed requests. As now __io_queue_sqe() is always used together with
struct io_comp_state, it leaves next propagation only a small window and
only for async reqs, that doesn't justify its existence.

Remove it, make __io_queue_sqe() to issue only a head request. It
simplifies the code and will allow other optimisations.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>