- Real => Float64 for better performance (Real is an abstract type)
- Should not create a type inside the model because the compiler fails
at estimating the type of the output of the f function (detected with
@code_warntype)
- Intentionaly ER is not as well improved as SIR in order to check the
differences of performance.

- Add of benchmark scripts that compares perf wrt pygmalion.
=> As expected it is bad in terms of reading cost (because of row by row
matrix read) but it is also not good in terms of simulation cost.
hcat seems to perform badly, should investigate.