speed up jobsub_history
jobsub_history calls condor_history and is agonizingly for both the user and the server. It is pretty much needed for users to check if their jobs have completed or died.
I did some testing that shows that properly formatted condor_history output stored in in an sqllite database is much much faster. Implement this with the following features:
- A process which periodically runs condor_history on the tail of the history log and puts these into the sqllite database
- All the jobsub_history (and jobsub_fetchlog file lookup) using this sqllite database
- Addtional functionality in jobsub_preen: when sandboxes removed, remove entries from sqllite database