Split 0.4 and 0.5 into separate classes

[alanocallaghan]

The current code tries to hide whether the underlying object is an 0.4 or 0.5 model by massaging away the differences between the 0.4 and 0.5 specs. For example, by handling Size vs Shape in tensor specifications.

However, there are some more fundamental differences in the data structures between the versions, eg attachments is a map in 0.4 and an array in 0.5.

Probably the most sensible thing to do is to implement verbatim representations of the bioimageio 0.4 and 0.5 specs as a baseline, and to then write code for common operations that handles the difference between 0.4 and 0.5 objects (eg, by overloading methods).

See #13 for some discussion on the details of the spec versions

This should happen after 0.6, because it may take some time.