Refactor check_submissions - adding API hooks for judging and minor multithreading improvements

[gkreitz]

This PR is a large refactor on the check_submissions method. Main technical changes:

• Added a class SubmissionsJudge which gets a call with a large batch of submissions to check. This is mainly to allow an API user to hook judging within problemtools, but also lets our multithreading schedule precompute a bit better.
• A complete rewrite of the logic in check_submissions. I feel like it reads a lot cleaner now (gather up all submissions which lower bound the time limit, run those, compute time limit, run the rest)
• Getting rid of directory hard coding (slowly making progress towards supporting submissions.yaml)

Behavioral changes:

• If a submission is larger than the code size limit, we now still compile and test it (more in line with other non-fatal errors)
• Minor change in the order we test submissions (alphabetical order, but forcing TLE last)
• "AC submission foo.java" is now "accepted/foo.java". This is arguable a bit worse, but when we eventually support submissions.yaml, submissions can come in arbitrary directories.