Including a version id in entity serializations should avoid any nasty surprises caused by breaking changes. Self-documenting data formats are a good idea, especially for formats that may potentially be used for long-term storage.
Some (but probably not all) questions that should be discussed in an RfC:
- location of the version number
- the version number should be in the dumps and in the API output; should it be somewhere else as well?
- should the version number be at more than one place?
- do we want different versions for different parts of the serialization?
- how could a mechanism for having multiple version numbers look like?