Asciidoctor already handles the caching so we don't really need to. On
the plus side, it gives some choice whether the pipeline should have a
cache or not.
This apparently not accepted since I think what happens is the
processing is taking place before whatever the class is initialized or
isn't considered by the text processor.