I have explored many different ways of building this library to make it easy to extend it for custom SQL dialects. Most of my attempts ended in failure but I have now found a workable solution. It is not without downsides but this seems to be the most pragmatic solution.
The concept is simply to write a new parser that delegates to the ANSI parser so that as much as possible of the core functionality can be re-used.
I also plan on building in specific support for custom data types, where a lambda function can be parsed to the parser to parse data types.
For an example of this, see the DataFusion project.