attached is an updated patch that also works in debug mode.  to do this
i had to exclude all zero length (value = 0) PyLongObjects from the free
list.  i'm still not sure why, they all have the same underlying
allocation size.

simplifying it to only freelist one digit longs the only useful
differences are in the microbenchmark of -32767..32768.  otherwise
things are slightly slower.  pybench is identical and pystone drops a bit.

i'm closing this as not worth it as things are written.  if someone
wants to pick up the idea and play with freelists go ahead but this
doesn't need to be an open tracker issue.  there is room for improvement
but these patches aren't it.