Optimize tuple_slice() and make further improvements to list_slice()

and list.extend().  Factoring the inner loops to remove the constant
structure references and fixed offsets gives speedups ranging from
20% to 30%.