feat: Add backoff support for retries.

[jimsynz]

This PR adds a new backoff/4 callback to the Reactor.Step behaviour that allows a step to dictate how long the executor should wait before rescheduling the step for execution. This is done by creating a new Reactor.Executor.Backoff struct which is added as a graph vertex as a dependency of the step being retried, which will stop the step from being executed until the dependency is removed.

The main executor loop sweeps the graph for expired backoff vertices and removes them only when there is not anything more useful for it to do, which means that the backoff times are a minimum delay and some jitter is inherent in the process.

Contributor checklist

Leave anything that you believe does not apply unchecked.

• I accept the AI Policy, or AI was not used in the creation of this PR.
• Bug fixes include regression tests
• Chores
• Documentation changes
• Features include unit/acceptance tests
• Refactoring
• Update dependencies