[ML] Categorizer state is not background persisted

[droberts195]

The intention is that we persist both anomaly detector state and categorizer state in the background periodically while a job is running (every 3-4 hours by default).

However, since version 7.4.0 the categorizer state has not been persisted in the background. Instead an error message like this is logged:

[2020-03-23T14:01:06,813][ERROR][org.elasticsearch.xpack.ml.process.logging.CppLogMessageHandler] [instance-0000000001] [my_job] [autodetect/15732] [CFieldDataTyper.cc@414] NULL persistence manager

The problem was introduced by the changes of #550. Prior to that the categorizer only needed a pointer to the background persister when it was unchained, i.e. running as part of a dedicated categorizer program. In #550 it was changed to use a new persistence manager class regardless of whether it was chained or not, but when chained (which is always the case in current shipped product) it was not passed a pointer to the persistence manager. Hence in production it never does background persistence. (The unit tests cannot detect this because the bug is in the main() function of a production program. It would need an integration test that uses the complete program to find the problem.)