Make summarization agnostic of schema ordering

[fharding1]

Currently schema ordering matters for computed properties. Either the way summaries are done should be updated to be agnostic of that, or we should use some sort of topological sort to sort the schema before we use it.