Currently, our worker definitions (including queue name) are defined in TOML.

However, the backend has the 'default' queue hardcoded for all tasks.

Maybe we could move that code into TOML.

Each job would have a map: into which queue it should go, and what the timeout should be.

So this kind of function: #505 would be parametrized by TOML.

If we had jobs in multiple queues, we could have a split for "fast" jobs (where 1-2 workers are enough) and "slow" ones. The slow jobs (even if queue is full of them) wouldn't block the fast ones.