Elasticsearch pending worker count is wrong on Nuxeo cluster with Redis

When using Nuxeo in cluster mode (with more than one node) with a Redis job queue the "Indexing worker pending" count is wrong.
Depending on where the job are executed it can give a negative number on one node and a positive in the other.

The status "Indexing in progress" or "No indexing in progress" is also corrupted.

The prepareWaitForIndexing api is also broken. This API is not used on a bare Nuxeo only during unit test.