Implement the Six Core Orchestration Patterns

PROJECT CONTEXT:
My multi-agent setup is ad-hoc: every coordination need gets a bespoke hack, and long-running agents leak resources or lose results. Claude Code's orchestration converges on six named patterns with shared infrastructure - lifecycle management, result delivery, and cleanup - underneath all of them.

GOAL:
Build the shared orchestration infrastructure (lifecycle, delivery, cleanup) and implement the six patterns on top: worker pool, pipeline, supervisor-worker, adversarial, fork-join, resume chain.

REQUIREMENTS:
- Shared lifecycle: register on dispatch, track foreground/background state, deliver results, clean up on completion
- Foreground-to-background promotion for long-running children without losing them
- Per-child transcripts persisted so completed or stopped agents can be resumed
- All six patterns implemented as reusable functions over the shared infrastructure
- Complete cleanup on every exit path - normal, error, and cancellation

CONSTRAINTS:
- Agent-to-agent communication must be explicit - a message not sent through the delivery path does not exist
- Every agent creation must have a matching cleanup or the system leaks under load

DEFINITION OF DONE:
- A long-running foreground child can be promoted to background mid-execution and still deliver
- Each of the six patterns runs end-to-end on a demo task
- Killing the parent mid-pattern leaves no orphaned processes, files, or connections
- A stopped agent resumes from its transcript and completes

COMMON FAILURES TO AVOID:
- Results delivered to a log nobody reads instead of the parent's next turn
- Background children killed accidentally by the parent's cancellation signal
- Cleanup on the happy path only, leaking sessions/files/connections on errors
- Implicit communication assumptions - agents 'reporting' in text no other agent receives