Nightly jobs: don't retry, longer timeouts, expectations updates

We don't want to retry nightly jobs, because they run for so very long, and it's not unexpected that they will fail. Make a618's full-run timeouts longer so that we do actually get to complete them more often. Update expectations for a630 and gc2000 based on the last nightly runs. a618 will no doubt have changes too, but we need to see completed jobs to get them.

Verified by looking at some jobs in the Rails console by hand that the full jobs indeed won't get retried but for infra reasons, and the other jobs still will.