docs: add accepted M9 professional runtime design

[shenjiecode]

Summary

• add the accepted M9 design for trusted execution, atomic Result submission, professional Agent Packages, project memory, lexical retrieval, and Task diagnostics
• split Task-scoped member execution from Work-scoped Leader coordination
• define PostgreSQL-owned execution claims, writer fencing, phased cancellation recovery, bounded ToolResult ingest, and the 22-item M9-A stop line
• define Package-private Skills, scope-terminal SkillUse interruption, Worker-owned professional validation, and generic Coordination settlement checks
• define RuntimeCapabilityBinding, ProjectBinding activation, static deployment/Team quotas, retention, tokenizer, and remote transcript boundaries

Verification

• Markdown heading and fence checks
• local links resolve
• all JSON examples parse
• no trailing whitespace, private paths, credentials, or private dependencies

Scope

Documentation only. This PR does not enable any M9 runtime contract and does not modify the canonical product designs. Canonical synchronization remains required before the corresponding implementation is enabled.

Follow-up

Run the M9-A0 pinned OpenClaw 2026.4.14 hook/bootstrap spike in a separate issue and PR before implementing Agent Packages, Skills, or retrieval.