test: validate the generated code by running a very long test

A number of bugs are hard to detect with simple unit tests. We should run a long test to shake them out, this may catch:

- Leaks: in the form of growing data structures or consuming limited OS resources
- Races and other undefined behavior: Swift makes this hard, but not impossible
- Failures to refresh the access tokens

The idea is to run a process that performs random-ish operations against production. Run this process until it completes 1 billion operations. If we run this on GCE, it is easy-ish to track memory usage, CPU usage, and other metrics.

The easies service to use is secret manager, it has a rate limit of about 1,000 requests per second (per project). That is about 100 million requests a day, so it may take up to a week to collect all the data. Allow some time for failed runs and to fix any problems uncovered in this run.