`convert` from `AbstractSparsePolynomialZonotope` to `Zonotope`

[schillic]

1. If there are no dependent generators, the implementation is trivial.
2. In 1D, the implementation is simple (low and high probably need to be improved in that case). (probably too complex, see below)
3. If there is only one parameter of degree 1, the implementation should still be simple.
4. All other cases do not seem to be a zonotope and should throw an error.

The first implementation could only handle the first case.

(The change in #3972 should use this convert method.)

[alecarraro]

I think the implementation of low and high can be quite tough. For the dependent generators you need to solve, for each coordinate, a maximization of a multivariate polynomial over a box domain.

[schillic]

True, it may not be that straightforward. But the true result is an interval and that can be represented as a zonotope.
In higher dimensions, I suspect that no matter how the dependent generators look like, the result should not be a zonotope; I guess for polynomials of degree > 1 and > 1 parameters...

The first implementation could just handle the first case, which is the most relevant one...