Abstract over the NodeState

[milibopp]

Currently NodeState is coupled tightly to its particular kind of memory layout:

pub enum NodeState<O, N> {
    Empty,
    Leaf(O),
    Branch(Vec<N>),
}

As the Node trait requires this exact type, it would not be possible for an implementation to use a different internal representation, e.g. [N, ..8] instead of Vec<N> for a dimension-specific octree.

Another problem is, that the queries are currently not implemented generically, which would be possible, if the state was generic, too. In principle, we could have impls like:

impl<T, P, N, O> ObjectQuery<O> for T
    where T: Node<P, N, O>
{ /* ... */ }

impl<T, P, N, O, D> DataQuery<D> for T
    where T: Node<P, N, O> + AssociatedData<D>
{ /* ... */ }

This would remove part of the code duplication between PureNTree and NTree.