Previously, every time we ran tests in parallel we would load the schema
for all N DBs.

Now that we have the ability to sync DBs with the schema file exactly
using metadata and a SHA, we can instead only load the schemas when they
change.