Split object store and data lake consistently

We should move to where:
* "object store" is a bucket in S3 where all the code (`schemas`, `jars`, ...) and configuration (`config`) lives
* "data lake" is a bucket in S3 where all (transient) data lives
    * `data` which is exported from upstream databases
    * data that is landed by other systems and is pulled in through Redshift Spectrum