job-archive interface: wrap job usage updates into a single SQL transaction

[cmoussa1]

Problem

When flux-accounting needs to do a job usage update for all of the associations in the association_table, it needs to make a number of updates to a number of tables in the DB. If there are a lot of associations, this can slow down the performance of a number of other, much simpler commands, such as flux account view-user. This is because the job usage updates makes a separate transaction for each UPDATE statement performed, forcing other reads from the DB to wait until the updates are completed.

---

This PR drops the separate commit() calls from the various helper functions that execute UPDATE statements in job_archive_interface.py and instead wraps the entire set of database updates into a single SQLite transaction, which will only be committed to the DB after all of the updates are done. This should hopefully begin to relieve some of the headache mentioned in #451.

[cmoussa1]

Thanks for reviewing @ryanday36 - the sharness test added in t1011-job-archive-interface.t is in fact checking job usage values for correctness. I was also able to confirm that wrapping up the updates into a single transaction is OK for calculating both user and bank usage values - subsequent updates made into a single transaction should be readable by each other. 👍 I'll set MWP here