After a crash, the "running" worker lists are not cleaned when using Redis

When the workmanager uses Redis, stopping brutally a Nuxeo instance while there are running jobs will not clean the work status stored in Redis.

As a result when restarting the Nuxeo instance the old workers are listed as running.

If these jobs are ElasticsearchIndexing work this will mess up the Elasticsearch indexing status on the Admin center.
It will also mess up the waitForCompletion method that count scheduled and running jobs.

A workaround is to flush the Redis database when the cluster is down.