Job dependencies: run job B only after job A completes

## Context

A common pattern is chaining jobs: "after the PDF is generated, send the email." Currently this requires the handler to manually insert the next job on completion. A declarative dependency mechanism would be cleaner.

## Design Options

1. **`depends_on` column** — job B has `depends_on = [job_A_id]`. Maintenance promotes B to available only when all dependencies are completed. Simple but requires tracking dependency state.

2. **Completion callback** — `InsertOpts { on_complete: Some(next_job_params) }`. The runtime automatically inserts the next job when the current one completes. Simpler but limited to single-step chains.

3. **Explicit "not a workflow engine"** — document that users should insert the next job in their handler. This is the current approach and keeps the system simple.

## Scope

- [ ] Evaluate whether this belongs in awa or should be left to application code
- [ ] If yes: design the dependency model (column? callback? separate table?)
- [ ] Consider: what happens when a dependency fails? Block dependents? Cancel them?
- [ ] Migration if needed
- [ ] Rust + Python API
- [ ] Tests

## Note

The PRD explicitly lists "workflow engine" as a non-goal. This should stay minimal — simple A→B chains, not DAGs.