Hi!

I've made a patch trying to speed up the interpreter a bit.

Essentially I've added two special nil classes for octave_base_value and idx_vector_rep that are constructed with an extra count.

In that way, some checks for equality with the nil object is unnecessary in some dtors that currently are afraid of calling delete on the nil object (which is not allocated on the heap).

Also, I moved the nil_rep getter function to the header together with the object itself, to allow for inlining in other translational units.

function i = testspeed ()

        tic;
        for i = 1:10e6
                i = 3* i / 2 + 1 - 88;
        end
        toc;

        tic;
        for i = 1:1000000
                qwe = [1:2000]; %idx_vector_rep
                qwe (3) = 43;
                qwe = qwe + 1;
        end
        toc;
end

For code like that, there seem to be a ~4% speed increase.