[michaelkedar]

(Sorry this PR is so big, a lot of things ended up having to be done at the same time)
Set up dual-writing infrastructure for Postgres in the places where vulns get written:

• Implemented the remaining VulnerabilityStore for postgres (splitting the queries.sql into files corresponding to the go files they're used in)
• Propagated data needed from importer (and recoverer) to worker to enable raw_vulnerabilities and historic_vulnerabilities to be written with all needed fields
  • Doing this meant passing a compressed JSON object instead of a compressed proto over the Pub/Sub for the raw vulnerability. We need to make sure there are no jobs in Pub/Sub while this is being merged (i.e. pause the importers).
• Implemented Dual-write wrapper, to write to Datastore/GCS, then Postgres. Using this for workers and vanir signatures.
• Updated relations cron-job to also mutate in Postgres (this wasn't using the database interfaces, and I have plans to change how this is done when postgres becomes the source of truth)
• Configured all the GKE images with the POSTGRES_* env vars (I've already set up the GKE secret for the db password)
• tests and such

I may have to make a follow-up PR fairly quickly to bump the CPUs of the postgres if affected commits are too plentiful.

In a future PR, I plan to re-jig the tests so they spin up postgres in a docker container instead of trying to mock all the endpoints, so we can make sure our queries actually work properly.