stop/resume jobs

[c-cube]

• when starting a job, put its full serialized description into a dotfile, and append each individual result as we go (use cbor+decoders?)
• command to list interrupted jobs
• command to resume an interrupted job (subtracting already done individual tasks)

edit: in xdg rundir, save a description of the job (a flat json file) and append results as we go. remove file when we're done.
resume looks for such files and picks up where ti stopped.

[c-cube]

after #23 :

• create a database as soon as we start the bench
• use a table for "tasks", use it as a queue for workers (with transactional get-and-remove? see what redis does). Tasks include:
  • discover the expected status of a file, see parse expected status as we go? #10
  • run one prover on one file
  • run a checker on a proof produced by a prover
• if we kill the benchmark system, it just resumes from the list of tasks that resumes

In a way all the state is in the sqlite db from the start.