Randomisations - add sample count randomisation

[shawnlaffan]

It would be useful to randomise the sample counts. The simplest approach is to hold the label ranges constant and re-assign the sample counts within those ranges, or perhaps completely randomly across all label/group combinations.

A more complex approach would be to run a structured randomisation where the total abundance of each group is held constant, but then any range constraint would need to be relaxed. There is also the risk that we would end up with landscapes of approximately equal richness scores in many cases, so possibly this would be best done as an extension to the spatially structured approach in issue #76