Revisit simplification of lazy Intersection

[schillic]

In #2482, we changed that Intersection with a HalfSpace argument automatically simplifies, e.g., to HPolyhedron. While this sounded reasonable, it breaks the assumption that Intersection always results in an Intersection object. The library has at least one place where this is assumed:

LazySets.jl/src/LazyOperations/Rectification.jl

Lines 535 to 537 in 264de96

	cap = Intersection(R.X, polyhedron)
	if filter_empty_sets
	set_isempty!(cap, false) # intersection is known to not be empty

Here, cap can be an HPolyhedron instead of the expected Intersection.

julia> rectify(rand(HalfSpace))
ERROR: MethodError: no method matching set_isempty!(::HPolyhedron{Float64, Vector{Float64}}, ::Bool)

Should we undo this change and rather leave this behavior to the concrete intersection operation?