feat: initial multi-objective support

[eddiebergman]

Adds multi-objective support to the NePS backend, meaning that users can now return multiple objectives and the runtime can handle it properly. NOTE No optimizers other than random or grid-search can actually handle multi-objective at the moment, this is only such that the infrastructure of NePS can handle it.

To return multiple objectives:

def evaluate_pipeline(...):
    return [y1, y2]

def evaluate_pipeline(...):
    return { "objective_to_minimize": [y1, y2], ...}

---

Since we never concretely define what a user can return, there is now a very dedicated parsing function which is pretty exhaustive about it and comes with a lot of tests. UserResult.parse(whatever_the_user_returns_from_evaluate_pipeline)

---

I also had to go through quite a bit of the summary related things to account for this. I also discovered CLI and plotting stuff was broken from before, and consequently, is also still broken during this change. This is a further thing that needs to be done.

We should consider if we really want to provide plotting code. It's usually quite custom per use-case, what's better is to provide examples on how to do it, but also just a way to export dataframes easily which people can interpret and plot as they like. This is now done by full_frame, short_summary = neps.status(root_directory), where full_frame contains all the information relating to trials, while short summary is more for printing purposes and is just a pd.Series.