C: Allow Arena to be reused across multiple `Herb.parse` calls

[marcoroth]

The way we integrate our Arena Allocator in #726 creates and destroys the arena for every Herb.parse call. Ideally we could find a way to allocate an arena and allow to pass it to any of the Herb.* calls from Ruby/JavaScript so the same arena can be reused across multiple calls.

[citizen428]

I think there's a design decision to make here: do you expect to reset the arena between repeated parses? That would invalidate previous results, which needs to be clearly communicated to callers. Or do you just want to keep growing it, which is probably also not ideal.

[marcoroth]

That would invalidate previous results

Maybe I'm not fully thinking this through, but I think it wouldn't invalidate the results, as we are copying/creating new objects using the Ruby C/Emscripten/NAPI API when we pass the parse/lex result from C to the respective environment. But there's also a good chance I'm missing something here.

I opened #1203 to explore this approach.

[citizen428]

Maybe I'm not fully thinking this through, but I think it wouldn't invalidate the results, as we are copying/creating new objects using the Ruby C/Emscripten/NAPI API when we pass the parse/lex result from C to the respective environment.

Yeah, I was admittedly thinking from a C-only perspective. You need to make sure not to keep references to reset pages around. So if you're creating new objects from that memory, you should indeed be fine.