[ML-DataFrame] Combine task_state and indexer_state in stats

[droberts195]

#43767 moved indexer_state into checkpointing.next, and it has been pointed out that this means it is only available when a checkpoint is in progress.

From an end user perspective the difference between task_state and indexer_state is an internal implementation detail. But for debugging purposes we might want to see it even when there isn't a checkpoint in progress. If we move it to the top level then as an end user I'm back to wondering which of task_state and indexer_state I should be taking notice of, and why there are two states in the first place. A better alternative is to have just one top level state that combines the two, like anomaly detection jobs and datafeeds have. It can be defined as:

• failed if what's currently reported as task_state is failed
• stopped if there is no persistent task
• Otherwise what's currently reported as indexer_state

To avoid multiple breaking changes to the stats format in consecutive versions and complex BWC this change should be made for 7.4.

[elasticmachine]

Pinging @elastic/ml-core