merge the legacy "mlp mean" project

[jacksalici]

As a second toy problem, it would be interesting to add an old project of mine, an MLP that shows how it can "learn" to mean operation. The parameters become all equal in a few iterations.