parallelise-OM

[tph-thuering]

Requirements

1. work out how to create multiple, independent, high-quality RNGs from a
   single seed
2. split population into several groups or assign each human to one of
   several RNGs
3. parallelise Population::update1

From another ticket

Potential addition with two objectives:
 * enable easy parallelisation of the human->update() call preserving
 repeatable results

Each human has an RNG state object (3-4 bytes), for independant random
number generation. One RNG state for non-human random calls.

Some initialiser for these generators (possibly another random number
generator):

*   initialised from seed in scenario
*   when humans are not created in initialisation (when previously would
    have been), use anyway to preserve old seeds

This initialiser could be:

*   another random number generator. (If using the same algorithm): some
    conversion between state of this generator and state used as seed for
    each created generator needed to avoid duplicating values
*   an incremented number. Not ideal because this is the same strategy used
    to provide seeds for scenarios.

Most suitible random number generators: taus, taus113, kiss

These are fast and only require a small amount of state information (see
http://home.netsurf.de/wolfgang.ehrhardt/misc_en.html, gsl source).

[dhardy]

Superceded by #204 and #194