[CM] pre-parsing scheme code

[bil at ccrma.Stanford.EDU]

If you're running the same code many times, it
might be worth it to call s7_optimize, save
the function it returns, then call that function
rather than eval+expression-writing code.  There's
an example in snd-sig.c.  The idea there is that
you're running a function over every sample of
some sound, so you first optimize it, then apply
that to each sample. snd-sig.c uses s7_set_slot
and so on to change the (user-level) function's
argument to be each new sample.  I should write a
simpler example...  (It's faster because you don't
rebuild the code each time you use it,
and the optimized evaluation is usually at least
twice as fast.)