I dunno that this will help much, it adds a lot of complexity,
but for my simple test cases, I see no major benefit.
It might be needed with more defined functions, so I'm saving it
here.
By not cloning code for recursion directly into the processing
buffer and by not copying as many strings and by not doing
string->int conversions all the time, it's significantly faster.
Specifically, it's about 5.5x slower than Python.
The idea I have is that I'll translate/compile JavaScript
into a rudimentary Forth like language. This sketch forms
the basis for that language's engine.