Add mechanism to prune `lis_result_sourcedid` table

Originally part of https://github.com/hypothesis/lms/issues/946

As currently designed, the `lis_result_sourcedid` table will have a row added for every student-assignment-consumerkey-contextId launch combination (for Blackboard users, at least), which could grow unwieldy-long over time if we don't garden it.

Originally pruning it was part of implementation, but after discussion, we've decided to break it out. The thinking was: this task should be managed by something (celery worker, cron, something) and not synchronous to web requests. Also: we don't need to (or, better-put: _can't_ really, without endangering functionality) start pruning the table for several months or a year.

Thus: once we have a chance to consider how to manage scheduled tasks on `lms`, come back and get this taken care of. The original thinking was to prune records that haven't been updated in some arbitrarily-long time, like, a year?