WIP: arena allocator improvements

Opening this can of worms with the following: Implementing h_arena_malloc() is needed, at least, if we want to compile a context-free parser from a h_bind continuation, at least when that parser includes the use of h_sequence(). The latter uses mm__->realloc and inside h_bind that is actually backed by a temporary HArena. Note that the provided implementation is known horrible, but a better one is not possible currently and at least it's not a failing assert().

I have a number of ideas on how to improve things in the arena allocator, so I'm making this a WIP area for future work. Feel free to merge along, though.