Split spec in two parts: isomorphic & lossy

[niquola]

It would be nice to make lossy transformations mandatory and separated in the spec, because from storage point of view we do not want to loose any information in database but for analytical goals it's ok.

[rbrush]

The spec as described on this page does not provide a lossless mapping -- since we don't support extensions on primitives, nested resources, or recursive types of unbounded depth. So if we need a lossless storage spec then that might be an entirely different document, perhaps?

For now I had simply limited my efforts to a (lossy) view of the underlying data that is easily consumed at scale with SQL...so the lossless model isn't a subject I've tackled here.