Large schemas take a long time to compile. If the schema is being compiled for every test it is very painful. One large schema has 110 files and takes 35 seconds to compile.

The TDML runner needs a way to load a saved parser, and in fact, to load it once, and not reload it for every test, but reuse it from test to test.

Obviously, flags to turn this off, for testing the save/restore feature itself, etc. are needed.