Add cache to Intersection

[schillic]

In #396 we implement the concrete intersection of two hyperrectangles. The issue with the lazy operation (Intersection{N, <:AbstractHyperrectangle, <:AbstractHyperrectangle}) is this: It needs to know whether the intersection is empty in advance. In the worst case this requires to compute the whole intersection, which to some extent breaks the point of the lazy operation.
Given that we have to compute this information with the first support vector query (or other queries like isempty), we should cache this result and only compute it once.
I guess that this observation is general for the intersection (in particular also for #375 and #262), so we should have a general caching mechanism for the lazy Intersection type. I cannot quite foresee how this cache would look like. For the moment it would suffice to have a new field isempty::Tribool. A Tribool would be a (mutable) type that can be true, false, or unknown. The field isempty would be initialized to unknown and later be set to true or false accordingly. (Note that Intersection would still be immutable.)
Later we may require more specific information for intersections, depending on the set types. Then we could replace the isempty field by a general caching field cache::C where C is a new type parameter of type IntersectionCache. The latter would be a new abstract type that we can instantiate for each respective combination of set types. In the instantiation we would store the respective information.

So the question is: Do we want to have the isempty field for now, or directly go for the general IntersectionCache type?

[mforets]

I like the idea of the isempty field (*). Besides or complementing this idea, what do you think if we ask that the lazy type at creation (it can be optional and true by default), it runs is_intersection_empty and in the case it is empty, it returns the EmptySet.

(*) not sure about this fields name, because this name is already taken..

[schillic]

(*) not sure about this fields name, because this name is already taken..

For a function only, right? The idea would be that we only use this field internally. And to me it describes perfectly what it should do.

what do you think if we ask that the lazy type at creation (it can be optional and true by default), it runs is_intersection_empty and in the case it is empty, it returns the EmptySet.

If it is optional, it is okay. But I hesitate to make it the default behavior because the is_intersection_empty check can be as expensive as the lazy version, which spoils everything. If you do not run this check, you at least save the computations if you never ask for a support vector - the only benefit I see with the Intersection wrapper.