Separate job inputs and outputs

The job's files/ endpoint allows to upload files necessary for computation, and download the results.

The main inconvenient is that it downloads everything that's contained in the folder, including the input data. It would thus be much more convenient to have the uploaded files put in a inputs/ subdirectory, and the downloaad enpoint return the outputs/ directory, if it exists.

(Another option could be to let the API fetch specific files, but it seems more complicated and not so useful)