Consider if the COG builder should only parse for certain subtrees of ontologies that CxG cares about

[nayib-jose-gloria]

i.e. don’t parse all of FBbt, just the nodes we care about per the schema). How to best bake that into our schema design and make it easy to update. Use schema to determine root nodes we need to care about in the cellxgene-schema validator + survey usage of COG in other downstream apps (for instance, validator doesn't need all CL terms but CellGuide does, so that one shouldn't be subtree'd)