Querying seems to be a fundamental factor of the problem. For instance:

- Upon canceling a want, canceling all wants it spawned, and the jobs attached to them.
- Answering the question, "what in-progress job runs were spawned by this want?"
- Answering, "why was this partition built?"
- Answering, "what partitions needed to be built and jobs run to fulfill this want?"
- Answering, "what jobs produce the partitions missed by this job run?"

Let's start prefixing functions that should probably be mostly queries with `query_`.


Notes on JTBDs and queries:

- When a want is schedulable (query), map the requested partitions to the job runs that create them (query), and start them
-