Versatile autograders

[inducer]

• Simply hand instructor code (from the repo) to a Python callable running in the container and expect it to produce a feedback JSON.
• Participant code could be one or more files, with instructor-defined languages.
• A lot of this code will be reused, so it'll need a way to do imports. (Maybe expose a subdirectory of the repo as importable by default?)
• Find a way to get the YAML from getting repetitive.
• A key in the YAML to supply data to the grading code