Improved Buchberger algorithm

[fredrik-johansson]

On the buchberger2 branch (commit a7caebe) is an attempt to implement the improved Buchberger algorithm GRÖBNERNEW2, p. 232 in Becker and Weispfenning. For some reason I can't figure out, it's not working as intended; random input quickly runs into cases where it slows down drastically while the ordinary Buchberger performs fine. Something like build/examples/dft -input 3 16 breaks.

At first I thought this was just due to randomness in the pair selection, but GRÖBNERNEW2 looks systematically worse, which shouldn't be the case. SymPy has an implementation of GRÖBNERNEW2 (https://github.com/sympy/sympy/blob/master/sympy/polys/groebnertools.py) and seems to do fine on a handful of examples I've tried manually.