Runtime: normalize `.aamf` log and output artifact directory structure

[jafreck]

Summary

AAMF writes useful observability artifacts today, but the runtime/agent/command logs and generated artifacts are spread across mixed top-level directories. This issue defines and implements a normalized directory layout for clearer operations, retention, and troubleshooting.

Problem

Current outputs are functional but harder to navigate at scale:

• Mixed log classes in a mostly flat logs/ layout.
• State files, runtime artifacts, and reports are not consistently grouped by purpose.
• Cleanup/retention operations are harder to target safely.

Goals

• Make runtime outputs predictable and self-describing.
• Separate state, logs, artifacts, and reports.
• Preserve backward compatibility for existing readers/scripts during migration.

Proposed Target Layout

Within .aamf/migration/{projectName}/:

• state/
  • checkpoint.json
  • checkpoint.backup.json
  • run-manifest.json (optional summary metadata)
• logs/runtime/
  • migration.log
• logs/agents/{agent}/{taskId}/
  • per-invocation stdout/stderr logs
• logs/commands/{build|test}/
  • command execution logs
• artifacts/contexts/
  • context JSON files
• artifacts/results/
  • structured sidecar result JSONs
• artifacts/planning/
  • planning outputs (groups.json, tasks-*.json, etc.)
• artifacts/parity/
  • parity/final parity reports
• artifacts/adjudication/
  • adjudication outputs
• reports/
  • progress.md
  • observability/ report outputs
• metrics/
  • invocations.jsonl, summary.json

Proposed Scope

1. Centralize path derivation in one helper/module to avoid drift.
2. Update runtime/orchestrator/context-builder/logger/agent-launcher to write to normalized paths.
3. Add temporary compatibility shims (read from old paths when new path missing).
4. Update README and fixtures/tests for new layout.
5. Ensure keepArtifacts and reset behaviors continue to work correctly.

Acceptance Criteria

• Runtime, agent, and command logs are written under dedicated logs/* subtrees.
• Checkpoint files move under state/ and continue to load/save atomically.
• Context and result sidecars are written/read from normalized artifact paths.
• Progress and observability reports remain generated and discoverable.
• Existing runs can still be resumed/read via compatibility fallback paths.
• README and tests reflect the new structure.

Test Plan

• Unit tests for path helper mappings.
• Integration tests verifying directory tree creation and expected file placement.
• Backward-compat tests: old-path artifacts still readable after upgrade.

Notes

This issue focuses on filesystem contract normalization only. Functional migration behavior should remain unchanged except for path locations.