feat: AI agents

[adityathebe]

resolves: #1719

NOTE: we don't want to trigger a child run and then wait for it's completion 
(especially with an open transaction).

Discussed approaches
- the llm tool simply trigger the run and goes into a "paused" state. 
  the child playbook "resumes" the action after its compleition.
- Alternatively, we trigger the run and add a new action to the run that
  waits for the child playbook's completion and continues it's execution.