io_uring: fix truncated async read/readv and write/writev retry

Ensure we keep the truncated value, if we did truncate it. If not, we
might read/write more than the registered buffer size.

Also for retry, ensure that we return the truncated mapped value for
the vectorized versions of the read/write commands.

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