Investigate quilt_rails DB connection pool exhaustion

[GoodForOneFare]

Overview

While running Genghis flows, Splunk shows that the Node SSR renderer receives many invalid GraphQL query responses (malformed JSON). While testing locally, I was able to reproduce these errors by:

• Increasing the Puma thread count to be greater than the database pool size
• Launching (puma_thread_count - 1) requests

Tackling this needs:

• A sample app that uses ActiveRecord connections
• Production deployment
• Genghis flows to exercise the production app

[GoodForOneFare]

@cartogram will take care of getting a public app that uses a database in #1055

[GoodForOneFare]

The production errors were almost certainly a symptom of controller deadlocks (#1059) causing query timeouts. That said, even the deadlockless "Rake hijacking" Genghis flows eventually ran into database timeouts, but only after serving 35,000 UI requests.

In development, it makes sense that having more HTTP worker threads than database connections causes errors. Rake just grabs a database connection for each request, so more workers than connections is an easy way to 💥