I have now seen GHA runs where we spend more than 24 hours for a testing workflow. The practical effect of this is that bundles that we current carry with `retention-days: 1` gets purged before jobs can use them. This then fails the test jobs that e.g. unable to pull jtreg.

This seems to happen in openjdk-bots -driven backports quite a bit (the run a lots of GHA runs, so they are often at capacity) and with Mac jobs (where the compute capacity is not great).

We can bump the retention time a bit to handle this a bit better.