[ML] Datafeed start time is incorrect when the job has trailing empty buckets

[dimitris-athanasiou]

Describe the bug:

When the user starts a datafeed, we pop a menu to guide the user to pick start/end time. By default, we have the option Continue from <timestamp> where is meant to be the point the datafeed has reached to (in the scenario the datafeed had run previously). This timestamp may be incorrect (seems to show the latest record timestamp) when the job contains empty buckets past the latest record timestamp.

Steps to reproduce:

1. Run a job in real-time
2. You just need a few data points to have a few buckets and then once it gets in real-time we want empty buckets
3. Stop the datafeed
4. Start datafeed to get the relevant menu and observe the issue

Expected behavior:

The earliest start time for the datafeed should be the max(latest_record_timestamp, latest_bucket.timestamp + bucket_span).

Screenshots (if relevant):

[elasticmachine]

Pinging @elastic/ml-ui (:ml)

[peteharverson]

Closing as fixed for 7.9.0 by #71976