Allow for more cache reuse at engine and translator level

To make better use of cache entries we should look at normalizing, allowing for subselection (fewer columns, additional predicates), and incremental approaches.