Hi!

I have made a patch that adds an object pool for octave_base_scalar and
its derived classes, to decrease malloc and free usage.

Code like:

function i = foo ()
        tic;
        for i = 1:100000
                i = 1+2+3+4+5+6+7;
        end
        toc;
end

has a speed increase of about 13% on my machine and the amount of mallocs
decreases from 1,1 million to 440 000.

The patch works fine on my x64 Debian machine, but should probably be tested
on a Windows build, because C++ thread_local support is kinda quirky.

The object pool size is set to 8, but could probably be tuned somewhat.