An idea that came up at work when implementing SpiceDB in a small K8s deployment - rather than have to push all our relationship updates to the backend (and update all the callsites) it'd be easier if we could use it as a read-through cache. I was thinking through a proposal where you can expose a "DataFetcher" interface for any

<resource, relation>

tuple that would get invoked on cache miss to update SpiceDB. This makes it easier to roll out, since our team can wire up all the fetching hooks to various stateful services without needing to integrate with them directly.