Restructure output format

[jonatanklosko]

I've been meaning to do this for a while, but with recent text changes it felt like a good moment.

So far we used tuples for outputs, but that is a bit annoying in certain places and makes pattern matching more verbose if we ignore tuple elements. Given that outputs are data structures that we keep, I should've done it from the start :)

I also moved output types specification from kino here. It's a fairly low-level detail, that doesn't matter for kino users in practice, and it is a part of the runtime specification, so here's a better place. Additionally, here we actually work with the outputs, so detailed typespecs can be more helpful.

I added a normalization function (Session.normalize_runtime_output/1) for backward compatibility and it handles all attributes that were added incrementally and may not be present. Going forward we should always normalize outputs there, so that we can assume complete outputs everywhere else.

[github-actions]

Uffizzi Preview deployment-34257 was deleted.