Update the analysis

The leaderboard's contamination verdicts come from a Docent rubric (see Trajectory analysis). To refresh it after adding runs:

1. You need access to the RustMizan Docent collection.
2. Upload the new .eval files to the collection and run the rubric on them.
3. Fetch the verdicts and commit the result:

   python fetch_docent.py   # needs DOCENT_API_KEY (or a DOCENT_TOKEN file)
   

   This writes data/docent.json; commit it and open a pull request.