I could not find the steps to prevent Metal3 from booting up a node when a customer needs to shut it down for maintenance reasons.

As a workaround, and after disabling Stonith resource for that node, we used step 10.2.4 [1] from 
https://docs.redhat.com/en/documentation/red_hat_openstack_platform/16.2/html-single/rhosp_director_operator_for_openshift_container_platform/index#proc_removing-compute-nodes-from-your-overcloud-with-the-director-operator_assembly_scaling-compute-nodes-with-director-operator
If this is the intended way, it would be good to add a section where we explain this step is required before doing maintenance. And also what step is needed once maintenance is completed.
 
The same is also true for Openstack 17.1 using director operator doc: https://docs.redhat.com/en/documentation/red_hat_openstack_platform/17.1/html-single/deploying_an_overcloud_in_a_red_hat_openshift_container_platform_cluster_with_director_operator/index#proc_removing-compute-nodes-from-your-overcloud-with-director-operator_scaling-compute-nodes-with-OSPdO
 
[1] Annotate the bare-metal node to prevent Metal3 from starting the node:
$ oc annotate baremetalhost <node> baremetalhost.metal3.io/detached=true
$ oc logs --since=1h <metal3-pod> metal3-baremetal-operator | grep -i detach
$ oc get baremetalhost <node> -o json | jq .status.operationalStatus
"detached"