convert from Taylor model to SparsePolynomialZonotope

[schillic]

The unit test is not perfect because it checks for equality, not equivalence; but we do not have an equivalence check for polynomial zonotopes. So I added another integration test that checks equivalence of the overapproximations, which in this case holds (I doubt it holds in general, though, so this is also not really a good test).
With #3446, the zonotope overapproximations are indeed identical (without #3446, they are still equivalent, but with lots of redundant generators).