Include model instantiation in benchmarks

[Tortar]

I saw that the model instantiation is considered a part of the setup of the benchmarks, but for more accurate comparisons we should instead include it in the benchmark itself. I could easily do this without harm for Agents.jl and Mesa, but can't say the same for the other frameworks :Q Don't know if there was some difficulty when this repo was created in the standardization of the comparisons between frameworks doing the benchmarks this way. At the same time this is a really minor issue probably since the setup probably doesn't cost a lot in all frameworks, will do some experiment including it in Mesa and Agents,jl to see by how much comparisons change if included.

Probably it's not worth the effort.

[Tortar]

Verified on the schelling and flocking models: the times are rather stable so I think that the time taken by the model instantiation is negligible.