Cache session active_at in Redis for periodic job processor

[jace]

This branch introduces two new patterns that need some discussion and consensus on:

1. Models that store frequently changing data (like active_at timestamps) in Redis, for periodic job processors to flush to database. Should we override the column property to transparently write to Redis instead of custom code each time? Should the flush to db have a standardised name like say flush_from_redis?
2. Models that use UUID keys and must expect queries with either UUID or string parameters. Where exactly is type safety enforced?