Add specification of input file format

[KasperFyhn]

Both issues #1 and #2 stem from problems with the format of the input file for supervised_classification.py, as there are some assumptions that are not entirely clear before running the script and digging into it, namely that: the positive label of a "label column" should be the same as the name of that label column. Additionally, if the script should be runnable, the input file most contain the columns raw_text, exemplar and political.

You partly address some of these things already in the README, but it wasn't fully clear to me, as is evident from the issues I have opened. 😄

So, specifying the format of an input file and/or giving an example file would definitely help.

This issue supersedes #1 and #2.

[x-tabdeveloping]

I think the solution would ultimately be to let users pass the name of the positive label to the script, because it is a pretty wild assumption to make that the column and the positive label are the same, especially if it isn't binary classification (which is what it happens to be, but then nowhere in the documentation do we say that this is necessarily the case).
This issue mostly concerns evaluation code, which was written by @miscodisco. Should we fix that up, or just document our way out of it?