Improve PDB cluster checks

[sebgl]

When updating the default PDB MinAvailable value, we look at:

• cluster health
• number of master nodes
• number of data nodes

It gives a good best-effort sense on whether 1 pod can be disrupted or not, but it has some flaws:

1. The cluster health information we use here is out-of-date, and comes from the last observed cluster health from Elasticsearch. We may see a yellow cluster as green, temporarily.

2. Even though Pods are reported Ready, they might not be part of the cluster yet. Which could lead us to allow one pod to be disrupted (7/7 Pods ready) even if only half of them are actually part of the cluster. We could check if all nodes belong to the cluster (_cat/nodes) before allowing one to be disrupted.