history-db may not commit to cache if killed early

Mostly we need to verify how this is going to work.

The specific concern is that initial import for a large branch might timeout, causing the thread to get reaped. If the import process hasn't done any committing, then the next request will have the same problem.
If we commit periodically during the cache update, we'll at least progress towards completion. So that even if the first request fails to complete, the second one can use the work that was already done.

Something like committing every 1000 revisions would probably be reasonable. As long as we commit the old history first.