Potentially "dangerous" changeset, please test

[Jordi Gutiérrez Hermoso]

I pushed the following which is suposed to optimise several cases for
randperm, which is important now that I've made randperm a central
function for the sprand family of functions:

    http://hg.savannah.gnu.org/hgweb/octave/rev/dd3c5325039c

However, this change is potentially "dangerous" because it uses an
"experimental" C++ feature (actually, it was experimental eight years
ago). If you see build failures on rand.cc related to tr1 or
unordered_map please blame me.

Thanks,
- Jordi G. H.