Bug in deep thought code

[fingolfin]

The following issue was discovered by my student, Nina Wagner. To reproduce it, you first need to set USE_COMBINATORIAL_COLLECTOR to true (this requires making it writeable), due to issue #9.

Apparently adding Hall polynomials can turn a confluent collector into a non-confluent one:

gap> z1 := [ 2, 0, 2, 1, 4, 5, 0, 3, 0 ];;
gap> H := PcGroupToPcpGroup(SmallGroup(512, 1002));;
gap> c := Collector(H);
<<from the left collector with 9 generators>>
gap> RelativeOrders(c);
[ 2, 2, 2, 2, 2, 2, 2, 2, 2 ]
gap> IsConfluent(c);
true
gap> PcpElementByExponents(c, z1);
g8*g9
gap> AddHallPolynomials(c);
gap> IsConfluent(c);
false
gap> PcpElementByExponents(c, z1);
g8^3

So in other words, the Deep thought code in GAP has a bug... Good thing it is completely undocumented and hence likely not used by anybody...