salt: fix unsafe approach for etcd `initial-cluster-state

Initially, we set etcd cluster state to new when the member
list from the pillars is empty.

This approach is not optimal since if for what ever reason we are
not able to retrieve the list.

Now, we raise an error when setting the cluster state to `existing`
if we are unable to obtain a list of members.
Also, we only set to `new` state only during a bootstrap.

salt/metalk8s/kubernetes/etcd/installed.sls

@@ -12,8 +12,12 @@ include:
 {#- Get the list of existing etcd member. #}
 {%- set etcd_members = pillar.metalk8s.etcd.members %}
 
-{#- Compute the initial state according to the existing list of node. #}
-{%- set state = "existing" if etcd_members else "new" %}
+{%- if pillar.is_bootstrap %}
+  {%- set state = 'new' %}
+{%- else %}
+  {%- set state = 'existing' if etcd_members else
+        raise ('etcd member list cannot be empty for existing state')  %}
+{%- endif %}
 
 {%- set etcd_endpoints = {} %}
 {#- NOTE: Filter out members with empty name as they are not started yet. #}

salt/metalk8s/orchestrate/bootstrap/init.sls

@@ -30,6 +30,7 @@
 
 {%- set pillar_data = {
         'bootstrap_id': pillar.bootstrap_id,
+        'is_bootstrap': True,
         'metalk8s': {
             'nodes': {
                 pillar.bootstrap_id: {