Introduce upgrade! for universal polynomials

[SoongNoonien]

As discussed in #2206 I've added a new upgrade! method which is better suited for the use in the universal polynomial code than upgrade was before.

Now one can even argue that the original upgrade method could be removed. I'm fine with that or leaving it as is. In #2198 things will change anyway.

[SoongNoonien]

I see where the problem is but I'm not sure if it is a problem with the change or if evaluate was broken before. The tests fail when a constant universal polynomial is evaluated at a univariate polynomial. They fail because the same universal polynomial gets evaluated prior to that which upgrades it. The upgraded polynomial is then internally a multivariate polynomial depending on three variables. And evaluate fails because it tries to map the univariate polynomial into the universal polynomial ring. Interestingly, evaluating the underlying multivariate polynomial at the same univariate polynomial fails too. So, I'm not sure if this evaluate which is tested here should work at all.