feat(recipes): persist parent_session_id in state.json for sub-recipe sessions

[samueljklee]

Problem

When a parent recipe invokes sub-recipes via type: recipe steps, the parent-child relationship is not persisted to disk. The parent_session_id parameter flows through the executor call stack (in executor.py) but is only used for cancellation propagation — it never reaches create_session() in session.py and is never written to state.json.

This means:

• No way to correlate parent and child recipe sessions after execution — dashboards, audit tools, and monitoring cannot reconstruct the recipe hierarchy
• Parent recipes appear "stalled" — when a parent kicks off long-running sub-recipes, it stops updating its own state.json. Time-based status detection (used by the recipe dashboard) marks the parent as stalled/idle even though its children are actively running
• No history of nested execution — for recipes that trigger multiple levels of sub-recipes (e.g., a discovery pipeline orchestrating code-tracer, behavior-observer, and synthesizer sub-recipes), there's no way to reconstruct what happened

Evidence

• session.py:97-136 — create_session() writes 8 fields to state.json; no parent reference
• executor.py:445 — parent_session_id parameter exists but is annotated "for cancellation checks"
• executor.py:2484 — _execute_recipe_step() passes parent_session_id to recursive execute_recipe() but never persists it
• Verified across 399 real state.json files on disk — zero contain parent_id or parent_session_id at the top level

Proposed Change (in amplifier-bundle-recipes)

1. session.py — Add parent_session_id: str | None = None parameter to create_session() and include "parent_session_id": parent_session_id in the state dict
2. executor.py — Pass parent_session_id through to create_session() at the call sites where sub-recipe sessions are created
3. Backward compatible — existing state.json files without the field continue to work; consumers treat missing/null parent_session_id as "no parent" (top-level recipe)

Downstream Impact

Once this field is persisted, downstream tools can:

• Build parent→child tree views of recipe execution
• Propagate status accurately (parent is "running" if any child is running)
• Show complete nested recipe hierarchies in dashboards
• Enable audit trails for complex multi-recipe workflows