Planner not considering all indexer nodes for placement of index partitions

Build : 6.5.0-4821
Last known good build : 6.5.0-4744

Some of the automated partitioned index tests are failing as the test expects the partitions to be spread across all eligible indexer nodes. But recently there is a change in the planner due to which the partitions are not spread across ALL eligible nodes, but only some. This can be easily reproduced with the following scenario:

1. Setup a 4 node cluster with index service on all 4 nodes.
2. Install beer-sample bucket and drop the primary index that gets created as part of the sample bucket.
3. So now we have 4 indexer nodes with no indexes.
4. Create a partitioned index : create index idx1 on `beer-sample`(name,city) partition by hash(meta().id)
5. This index should be created with 8 partitions spread across all 4 nodes since all 4 nodes should be considered good by the planner.

The index gets created with partitions spread across only 2 nodes. The other 2 nodes do not have any partitions for this index.

The tests are designed to validate that in such cases, all 4 nodes are eligible to receive some partitions. These are failing as the planner is not using all 4 nodes for placement of partitions.

This can be related to MB-36964