Refactored submodel system to use ModelSpecs

[jlstevens]

This PR fixes some annoyances with the submodel system:

• topo.sim.model.V1 does not refer to a declarative SheetSpec object - instead this points to a method called 'V1'. Although important for defining the model class, this isn't particularly useful to the user once the model object has been created. It is confusing to have model.V1 and model.sheets.V1 on the same object.
• The subclasses of Model were a mix of declarative objects (e.g. model.sheets and model.projections) and the decorated methods used to create those objects. In contrast, SheetSpec and ProjectionSpec existed as distinct declarative objects (which can be called to be instantiated).

The solution proposed here is to introduce a ModelSpec which is the declarative counterpart to a Model definition class. Just like SheetSpec and ProjectionSpec it can be called to instantiate the corresponding model. This helps make a clearer distinction between the model definition (code), the model specification (lightweight, declarative object) and simulation components (memory hungry objects).

[philippjfr]

Makes sense. A few comments:

• I think it makes sense to keep the Model specification separate from the model definition. It might also make it easier to eventually move toward a modular system, i.e. by passing the model spec around.
• Why does attrs need to be an AttrTree? A simple dict would eliminate the need to use upper-case names, which are a bit awkward.

[jlstevens]

Ok great!

I'll make this commit directly to keep the commit logs clean, making sure to include a reference to this PR. I also agree that attrs doesn't need to be an AttrTree - I will switch it to a regular dictionary in a later commit.

EDIT1: Applied these changes to master in commit e0db6188 EDIT2: Switched attrs from AttrTree to dictionary in commit 3fd64fa