I've factored things to bring them inline with my expectations, except
for core parsing logic, which will be a seperate undertaking.

Notably, it has zero dependencies. Not even on headers!

Rather than relying on standard library to allocate scratch memory from
the heap, we have users provide us some scratch memory. This removes the
only dependency on the standard library, and allows users to parse using
only the heap.

I've also included minor odds 'n' ends, like documentation.

The adjusted parser will land next.

Core of the parser is left mostly as-is so as not to introduce bugs.