`write_vectored_all_at` should use `write_vectored_at`

[Berrysoft]

Also for the read ones...

write_vectored_all_at currently uses write_at because we don't know how to slice a vectored buffer if it is not consumed totally.