Support "totally unobserved" relations

[ThomasColthurst]

A totally unobserved relation (either clean or dirty) is one that is neither directly observed nor is the base relation of a directly observed dirty relation nor is the base relation of a chain of dirty relations the final one of which is observed.

We should support these! Preferably in the core IRM/HIRM code. Why? Because it is very nice for our users to be able to take any existing schema file and run it on a subset of an existing observations file without having to worry about editing down the schema file.

(Or in the pclean context: running it on a subset of columns and/or a subset of query fields.)