materializations: don't persist a spec

[williamhbaker]

Apply and Validate RPCs now include the last applied / last validated spec, so we don't have to persist it in the endpoint. This would be a nice simplification for materialization connectors, and may alleviate & prevent future problems with spec persistence (being too big, etc.). So we should stop storing specs in metadata tables. Something to double check is how disabling and re-enabling bindings would work with the new way of getting specs: Currently the materialization completely forgets about the old binding with the persisted spec. How does that work with the ones that are present in the protocol messages?