Caching should be based on ontologyRid, not on Client

[ericanderson]

As we implement multiple ontology support, it is expected that syntax will be client(ObjectType, ontologyRid) or client.withOntology(ontologyRid)(ObjectType). In both cases, if we cache by Client then we need to also keep around the rid -> client mapping so we might as well just cache on rid.

[ericanderson]

I'm second guessing this now.

• Its probably best to just leave it connected to the client so consumers have the most control. For example, they could force a reload of data by replacing their client, something they would not be able to do if we did ontologyRid.
• Using a second ontology client is likely more natural than client(ObjectType, ontologyRid). Documentation could convey that you should hang onto client.withOntology() (if we build that syntax) but if you don't you just see perf issues.

I think we get the least amount of surprise and the most possibility to do something about it without this.