[Nickle]Performance tuning

[Keith Packard]

After waiting for nickle to draw anti-aliased polygons, I decided to go 
and spruce up the nickle run-time system a bit.  As is usual with such 
processes, careful tuning has improved the performance by a factor of 2.

We have a new performance metric now, NIPS -- nickle instructions per 
second.  My 1133MHz Athlon gets about 8.5M NIPS.  This is easy to measure
on your own, just use:

	$ time nickle
	> for (int i = 0; i < 10e6; i++);
	> quit

Each trip around the loop takes 6 instructions, so divide 60e6 by the
user time reported here.

Reports of trouble are welcome; things "should" be the same, but I'm sure 
I missed testing at least one code path that's changed.

-keith