(a)rc gets counted as 0 size

this is quite easily solvable by utilizing duplicate detection during traversal using a Set.

there are two ways to go about it: either change the signature of estimate_heap_size from (&self) -> usize to (&self, &mut State)
or rather create a new function that passes along said state and keep estimate_heap_size the same (it internally calls the new function).

or use a thread local, but i would like to avoid doing that.

either way is a breaking change, since people need to adjust their code.
i would be willing to create a pull request of the first variant if that is something the maintainer is interested in.