REST tests get run against either 1 node or multiple nodes. Wait for yellow with replicas>0 is not enough when running against multiple nodes as replicas shard might get initialized during testing, which can cause timing issues.
delete/50_refresh tests per shard refresh using refresh:true in delete api. We might run into troubles though if we have a replica that gets initialized after a doc was indexed and deleted, without a refresh, as that doc won't be found when searching against that specific replica shard (as a refresh happens automatically before a replica gets exposed as started).
This should make easier any potential expansion such as:
client.update('index', 'type', 41, {...}, ignore: 409
Also restructured and renamed the test suite a bit as part of the commit.
---
This reverts commit 0b142cc8812d8d090e48ea0ad89c007767870b91, "Removed client-specific `ignore_missing` parameter".