Add a section-ordering script

[kmod]

I think this lets us specify that certain functions should be
put together at the end of the text segment.

I picked some random functions that were showing up in the perf report
and listed those. I think there's a lot we can do here for improving
performance, but my main goal for now is to improve performance consistency.

              pyston django_migrate.py                  :    2.1s tp_hash: 2.2 (-1.1%)
              pyston virtualenv_bench.py                :    5.7s tp_hash: 5.8 (-1.1%)
              pyston django-template.py                 :   10.9s tp_hash: 11.4 (-3.8%)
              pyston interp2.py                         :    5.4s tp_hash: 5.5 (-1.7%)
              pyston raytrace.py                        :    6.5s tp_hash: 6.4 (+1.0%)
              pyston nbody.py                           :    8.4s tp_hash: 8.4 (+0.9%)
              pyston fannkuch.py                        :    7.3s tp_hash: 7.5 (-3.0%)
              pyston chaos.py                           :   20.5s tp_hash: 19.9 (+2.9%)
              pyston fasta.py                           :    5.4s tp_hash: 5.5 (-1.4%)
              pyston pidigits.py                        :    5.7s tp_hash: 5.7 (-0.9%)
              pyston richards.py                        :    2.4s tp_hash: 2.5 (-1.4%)
              pyston deltablue.py                       :    1.8s tp_hash: 1.8 (-1.5%)
              pyston (geomean-1a6f)                     :    5.5s tp_hash: 5.6 (-1.7%)

[kmod]

Ok I tried to be more careful with the building and measurement, and this is what I got:

              pyston django_migrate.py                  :    2.1s baseline: 2.2 (-2.3%)
              pyston virtualenv_bench.py                :    5.7s baseline: 5.7 (-1.3%)
              pyston django-template.py                 :   10.8s baseline: 11.1 (-2.4%)
              pyston interp2.py                         :    5.4s baseline: 5.4 (-0.7%)
              pyston raytrace.py                        :    6.4s baseline: 6.2 (+3.1%)
              pyston nbody.py                           :    8.4s baseline: 8.4 (+0.4%)
              pyston fannkuch.py                        :    7.2s baseline: 7.7 (-6.0%)
              pyston chaos.py                           :   20.2s baseline: 20.1 (+0.5%)
              pyston fasta.py                           :    5.4s baseline: 5.5 (-1.5%)
              pyston pidigits.py                        :    5.6s baseline: 5.7 (-1.3%)
              pyston richards.py                        :    2.5s baseline: 2.5 (-1.0%)
              pyston deltablue.py                       :    1.8s baseline: 1.8 (-1.2%)
              pyston (geomean-1a6f)                     :    5.5s baseline: 5.6 (-1.2%)

I messed with the ordering some more and got some different results.

I guess this gives a decent sense of how much text layout (which is all this change affects) can affect overall perf. It also says to me that our previous crop of benchmarks is not very good and motivates me to replace it -- our newer benchmarks show much more consistent and understandable changes.

But anyway, I'm going to push this for now, and see if it can help with the next set of perf investigations.