Add a timeout flag to agents

We are experiencing some issues where a job may sometimes hang and because of this our stack ends up not scaled down during the night.

We can add `timeout_in_minutes` in individual steps, but this being opt-in is not ideal. What we'd really want is to impose a hard global limit of the form "no job can run for more than XX minutes". Doing this at the agent level probably makes more sense, since we could then use different queues for different hard limits if we wanted to.

I see there was a PR trying to implement this (#788) but unfortunately it was never finished.

Start by reviewing the issue requirements and the unfinished PR #788 to understand the prior approach and remaining design questions. Define how an agent-level timeout should apply across jobs and how completion can be verified; done means a configurable hard limit prevents jobs from running beyond the selected number of minutes.

Written by the indexing model from the issue text.