Having found the nice docs at:

I have divined that I can ssh to each controller node and:

sudo docker exec -u root nova_scheduler crudini --set /etc/nova/nova.conf placement randomize_allocation_candidates true

sudo docker kill -s SIGHUP nova_scheduler

...and indeed the /etc/nova/nova.conf in each nova_scheduler container is updated accordingly.

Unfortunately, instances are all still launched on compute-0.