Hey Remi,
It's great to see someone else interested in adversarial stylometry! I have been working on a project quite similar to your proposal since January as my undergraduate thesis. I am considering applying for TSoP in order to fund the continued development of the project, but I think the mentor issue is a strong indication that we should focus our efforts on something more related to Tor. Since I have made considerable progress implementing the tool you described, perhaps we should instead combine our efforts? Rachel Greenstadt, whose work I am sure you are familiar with, also expressed interest in collaboration when I announced this project in December.
You will find that I have implemented many of your proposed features in my project on GitHub. Near-term work involves making the project easier to install and implementing Writeprint feature set extractors. If you manage to get it running (for now, you will have to install dependencies manually), I would love to hear some feedback from you! Until then, see my thesis.