database: look into other optimization techniques during a job usage update

[cmoussa1]

It looks like the flux account commands do not hang so badly anymore during a job usage update with the changes made to wrap all of the updates into a single transaction (it used to take more than a couple minutes and now it takes a little less than 8 seconds on a system with very high utilization). I should look to see if there are any other techniques I can use to further optimize the job usage update in addition to wrapping the UPDATE statements into a single transaction, in particular with any of the READ operations from the database.