Deterministic benchmarking for randomized games

[casper2002casper]

When game environments are randomized, some instances will result in larger rewards than others, resulting in noise in the benchmarks. Using enough games can average this out, but this is inefficient for games which are slow to solve. This PR adds a random generator to the init function, which can be used to construct the environment. As the seed of the RNG is using the sim_id the same environments will be generated for each iteration benchmark. Ideally, the implementation influences non-randomized games as little as possible.