vhost: avoid backend flush on vring ops

vring changes already do a flush internally where appropriate, so we do
not need a second flush.

It's currently not very expensive but a follow-up patch makes flush more
heavy-weight, so remove the extra flush here to avoid regressing
performance if call or kick fds are changed on data path.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>