[ML] Improve usability for AD model snapshot revert

[pheyos]

Reverting to an old AD snapshot deletes all anomalies back to that point in time.
This can potentially take a significant amount of time. And during that time the ML app stays in that "disabled" state without a loading indicator or a message warning about long execution time, which is not a great UX.

From my point of view going back to an old model snapshot is a bit similar to deleting a job (in terms of anomalies that need to be deleted in the background). I'd suggest we give it a similar user experience, like letting it run in the background and disabling other job interaction during that time.

[elasticmachine]

Pinging @elastic/ml-ui (:ml)

[jgowdyelastic]

From a request point of view, reverting a snapshot is quite different from deleting a job.
When we delete a job, a request is sent to es to delete the job, we receive an instant acknowledgment and a task is started in es to do the deleting in the background.
we then poll the current running tasks to know when to update the jobs list.

When we revert a snapshot, the request to es will wait until the revert has completed and so we keep the reverting modal open.

We could change the behaviour to close the modal as soon as the request is sent but the request would stay open in the background and we would show a toast once the request completes.
We'd also need to add a comment to the modal to say that the revert may take some time.