Make evaluation work with continuous logs analysis

[jlewi]

Per #84 and #139 we want to continuously process and ingest logs.
How do we update evaluation to work with this? There's a couple key questions

1. Where is the Analyzer actually running?
2. How does the evaluator know when the logs have been processed and can run steps that require them?

[jlewi]

The Analyzer needs to access the pebble databases. The pebble databases can't be multi-process. So if evaluation is reusing logs/databases used by the active instance of Foyle then the evaluation binary can't run its own Analyzer.

I think it makese sense to give each experiment its own set of logs and DBs so we don't pollute the logs/DBs of the production instance.

The experiment process could then run its own Analyzer.