[CT-2668] Improve ref() Parsing for SemanticModel Objects

[peterallenwebb]

As observed during review of our SemanticModels work, there are a couple important improvements which could be made to the way we are parsing the ref() expressions in SemanticModel.

• We should catch internal parsing exceptions and raise a more informative one.
• We should raise an exception if the ref does not resolve to a model.
• We should consider resolving refs at parse time, rather than deferring until the entire manifest is (almost) ready.
• We should check whether the ref'd model is disabled, and raise a specific exception if it is not.