[CT-1313] Investigate ways to improve performance of MacroGenerator calls in contexts

One of the performance bottlenecks for the macro calls that are inserted into every provider context to enable calling macros is that an entire set of macros is created for every model. In a previous effort to improve the performance of parsing tests we limited the macros in the test contexts to macros that actually were called in the code via some jinja parsing and improved the macro depends_on. However those changes were not implemented for models because it's much more difficult with models to know which macros are actually getting called.

Investigate ways of improving the performance in this area. Ideally we would be able to create a set of macros and reuse them by swapping in model references.