Make unit tests use redis

[cognifloyd]

The graceful_shutdown tests have been very flaky using the NoOpDriver. Switching from NoOpDriver to RedisDriver will hopefully make our CI more stable.

This was discussed in a TSC meeting and implemented by @FileMagic (#6223) and @guzzijones (#6236). I cherry-picked the redis-related parts of their excellent work. After stumbling through the cherry-picks (I missed a few things, and had to rebase/cherry-pick a few times), I refactored a few things:

• GHA: use the redis container defined in services, dropping the tasks that managed it manually.
• use ST2TESTS_REDIS_* as the format for new env vars instead of ST2_OVERRIDE_COORDINATOR_REDIS_*.
• fix a regression I added in #6241 that only became apparent once we stopped using NoOpDriver. It's surprising and scary how many of our tests rely on import-time side-effects.
• refactored tests to ensure each test got clean config. Several tests were relying on the config changes previous tests made, making them somewhat brittle. This will hopefully make that better.

:clap: Thank you @FileMagic and @guzzijones for getting this figured out! I have high hopes for more stable CI thanks to your excellent work! :tada:

[cognifloyd]

The Test / Test (pants runs: pytest) CI failures are not related to this PR. To validate that, I re-ran the success CI in #6244, and now it fails there too. So, it seem to be some GHA issue.

Luckily, the pants workflows are not required to merge PRs (in branch protection rules), so please ignore the failure.

[guzzijones]

Thanks for going through this. The noop driver random failures were a pain to deal with.