XSLT1 is too hard to work with, and Khronos have an annoying tendency to
generate their spec such that it's not in dependency order. So we're
switching to python, and performing basic depedency analysis within a
new tool.
This should allow us to do more extensive type modifications and
annotations.
