Even more graceful shutdown

[mariusandra]

When we deploy a new plugin server, we get errors like this: 1, 2.

These timeouts never get cleared because somehow the task itself ended early.

We should probably delay a bit more during shutdown?

[mariusandra]

Will investigate these issues as they come. Refactoring the kafka pipeline definitely silenced most of them though. Once we implement "recovery mode" as described here - https://github.com/PostHog/plugin-server/issues/353 - we will have a more graceful shutdown... as we'll need to boot up the recovery mode on normal SIGTERM shutdowns as well, to e.g. flush all memory buffers to persistent storage.