Improve shutdown behavior

Right now we mark all tasks as failed on shutdown.

Instead, pick one of:

- Serialize currently tracked jobs on disk, pick where we left off on restart.
- Wait for currently running jobs on shutdown.

The issue names no files, tests, or entry points. Start by locating the shutdown handling and how currently tracked jobs are represented, then determine whether persistence across restart or waiting for running jobs is the intended direction. Done means shutdown no longer marks every task as failed, with the chosen behavior covered by tests.

Written by the indexing model from the issue text.