Test against various common ES bad states

[temsa]

ES cluster are sensible to a number of issues that can happen during massive reindexations:

• cluster crashing
• no more memory stale
• cannot attribute a new shard
• cannot replicate a shard
• not enough vm.max_map_count
• split brain issues due to a network partition, generally leading to a loss of data
• shards that did not have time to come back from queries (e.g. 4 shards coming back from a search query instead of 5)

We should find some ways to test those states, and check how our create/update/delete operations perform in this case