Improve debuggability

[lawrencegripper]

Currently the dispatcher removes jobs (and hence their pods and the pods logs) when they have failed or completed.

This makes debugging a failed job very hard as the logs are logs. I

I propose moving to leave the jobs and then clean up any jobs that are older than 1 hour and have failed/completed.

Also the logging isn't great a lots of points and byte arrays are logged which aren't helpful, planning to clean this up too.

[lawrencegripper]

I've added some additional debug output to the committer as I'm seeing an issue where events appear to be lost between modules.

[lawrencegripper]

I'd like to add an additional doctype to the work in #135 that an be used by the ion trace command I mentioned in #122. This would create a document storing the stdout and stderr from the module runs. I propose that these types have a ttl so they automatically get deleted by cosmos.

This would allow ion trace to show eventMeta, insights and failures for a given correlation id in one place.