Random Network generation Improvements

[ptheywood]

• Random networks currently can include self-interactions
• Random networks currently can include duplicate interactions within a day
• Random networks are currently initialised from a truncated normal distribution per age group from model parameters. The reference model includes optional (bool paramater) use of negative binomeal.
• Random network generation is serial on the host. I.e. it will not scale well. It's doable in parallel, but not so much including the changes mentioned above
• The random generation is purely random, there is no spatial locallity. I.e. it is equally likely that someone in Swansea would randomly interact with someone in Bethesda. This could be improved, but is the same in the reference model.