Performance of partial differentiation

As mentioned in !62 (merged) and #95 (closed) it seems to be, that the current implementation of partial differentiation, especially of Newton polynomials, has some performance issues. For instance, the first micro-benchmarks (see here and the plot below), show, that the function partial_diff is noticeable slower for Newton polynomials compared to canonical ones. However, the overall scaling of the preformance along the number of polynomials seems promissing: