python code generation has a bug

[johnynek]

see the diff here:

d32020f

this shouldn't change anything: the order we match cases may change efficiency, but it shouldn't change the result in cases like this.

But inspecting the generated python code it is definitely wrong. The bug seems to be with python generation and not the Matchless AST since that is used for the interpretter tests which pass in both cases.

The cmp_BinNat expression is simple enough, but clearly miscompiled when the order isn't starting with Zero.