With respect to range_eq8_normalize2.patch, it is unusual to have a 
function that consumes a reference to its argument.  I would combine 
normalize_stop with PyNumber_Index and make it similar to validate_step 
with respect to reference counting.

Note that if you choose stop = start + len*step normaization, you will 
not need to create 'one' in normalize_stop.

With your patch I see

>>> range(0,6,2)
range(0, 6, 2)
>>> range(0,5,2)
range(0, 5, 2)

I would expect one of these ranges normalized.