Add option to reprocess runs in parallel

[takluyver]

To address #240.

This isn't particularly elegant code, but it does work, at least in a quick test. The most awkward bit is with the extractor object: I can't just use the same one across all threads, because at least the sqlite database connection shouldn't be shared across threads. But I also don't want to create a new Extractor for each individual run, so I didn't just map() over the available runs. I'm still thinking about how to handle this better.

I wanted to show less output when processing in parallel, because the usual output would probably be too fast and too jumbled up to be much use. The longer output is still directed to the relevant file in the process_logs folder.

[JamesWrigley]

For the record, I'm fine with creating extractors for each run :) That's essentially what's happening now when I use parallel anyway.

[takluyver]

Thanks, that's good to know. I vaguely remember that sometimes setting up the Kafka producer could take a while, so for recomputing one small variable in several runs it was an annoying overhead to recreate it each time.

I'll have a look (next week) if I can find a neater way to do it, otherwise I can fall back to just creating a new Extractor for each run.

[takluyver]

I'm going to close this, because I think #270 meets the need better, by submitting jobs to the solaris cluster to run in parallel.