Remove large serialisation inefficiency

[cortadocodes]

Current situation

Currently when we serialise Serialisable instances to a dictionary of python primatives, we go through a json.dumps call followed by a json.loads call with our custom encoder and decoder to ensure nested attributes that are e.g. datetime instances, sets or any of our own class instances with their own serialise methods are converted to a JSON-compatible form. Another reason we do this is to validate outbound information the same way as we validate incoming information. This is, however, slow and inefficient.

Proposed solution

A more performant method would be to implement an encoder which returns python primitives, not strings.