feat: implement multi-step planning loop with parallel execution and …

[peterish8]

📋 Summary

Implements the four-phase planning loop described in issue #405, giving autonomous agents a structured way to approach complex multi-step goals. The existing ReAct loop handles single-shot tasks but has no mechanism for breaking down a complex goal, tracking progress across steps, adjusting when something fails, or synthesising partial results into a final answer.

🔗 Related Issues

Closes #405

---

🧠 Context

Agents given multi-part research tasks or complex data transformation pipelines today have no structured approach beyond fitting everything into a single prompt. This change adds a proper planning abstraction at the kernel level and a concrete executor at the foundation level, following the existing microkernel architecture.

---

🛠️ Changes

• mofa-kernel: New Planner trait with PlanningRequest, ExecutionPlan, PlannedStep, MemoryItemSnapshot, and CompletedStep types. ExecutionPlan provides validate() and topological_groups() for dependency-aware parallel scheduling.
• mofa-foundation: PromptPlanner (LLM-backed Planner impl) and PlanningExecutor running four phases — planning, parallel execution, per-step reflection, and synthesis. Independent steps run concurrently via futures::join_all. Long-term memory is recalled during planning via the existing Memory trait.
• examples/planning_agent: End-to-end example — three parallel research subtasks feed a synthesis step that produces a structured report.

---

🧪 How you Tested

1. cargo check -p mofa-kernel -p mofa-foundation — passes clean
2. cargo test -p mofa-kernel planner — 6 tests pass (topological sort, cycle detection, validation edge cases)
3. cargo check -p planning_agent (from examples/) — compiles clean
4. Three foundation planning tests cover: parallel step execution timing, recalled memory in planning prompt, and reflection-triggered re-planning

Note: mofa-foundation coordination tests have pre-existing compile errors on main unrelated to this PR.

---

⚠️ Breaking Changes

• No breaking changes

---

🧹 Checklist

Code Quality

• Code follows Rust idioms and project conventions
• cargo fmt run
• cargo clippy passes without warnings

Testing

• Tests added/updated
• cargo test passes locally without any error

Documentation

• Public APIs documented
• README / docs updated (if needed)

PR Hygiene

• PR is small and focused (one logical change)
• Branch is up to date with main
• No unrelated commits
• Commit messages explain why, not only what