Can't capture batch name for jobs that have left the queue

[JoshKarpel]

The only place we can get guaranteed batch name information is when jobs are still in the queue, via a query. If you ask us to parse an event log whose jobs have all left the queue, we have to go with the default batch name. Although this is probably not a common use case, it will probably be surprising when it happens.

[JoshKarpel]

For now, we should probably just manage to emit a warning when this happens.

[JoshKarpel]

This gets particularly nasty when you're running a DAG with many completed jobs in it. You get thousands of rows, because we don't understand how to group them after they've the queue, and unlike a normal submit, they probably all have different cluster ids.