The population length of the `algorithms.Population` is not constant

[drvinceknight]

@GitToby noticed this:

It is possible for the population size to both increase and decrease.

The increase would happen only once and is essentially due to the self mutations (as opposed to the standard GA where all new entrants in to a population are from crossover). The decrease happens because this line: https://github.com/Axelrod-Python/axelrod-dojo/blob/master/src/axelrod_dojo/algorithms/genetic_algorithm.py#L109 should in fact be:

size_left = self.size - len(self.population)

I think that no matter what we should fix the decrease. (And add tests about the population size). My student is going to take a look at the performance of various variants (are the self mutations helpful?) and then we could potentially take a look at modifying the algorithm and/or adding a few more variants (in which case we need to document that the algorithm has these self mutations which for some choice of parameters can mean the first generation increases the population size).

[marcharper]

Good catch. Looks like we need to work on more tests in general for the dojo library.