Make cleanup threshold configurable per-resource group

As we run larger and larger scale tests, some tests may take longer to run than the 4h that is currently configured to allow extra time for incrementally scaling clusters up to their full size and collecting logs. It would be great if we could let those jobs run longer while still cleaning up other resource groups more eagerly. Something like a new tag that could be set to `8h` and used in place of the `--ttl` command line flag for a particular resource group could enable that.

We might want to be careful with this though so we don't make it possible for a malicious actor to start a job that runs their favorite crypto miner or whatever for an indefinite amount of time. `/ok-to-test` should protect against that for presubmit jobs, but maybe we should still have a fallback maximum.

(This isn't urgent, just something I thought of that we might need in the short-ish term.)