perf(moonep): optimize combine v2 ring scheduling

[SilentEmotion]

Summary

• optimize MoonEP Combine V2 ring/grant scheduling, WQE batching, and self-copy handling
• extend Combine V2 layout, schedule, source-guard, and hardware probe coverage
• add multi-host performance build, deployment, execution, aggregation tooling, and design documentation

Validation

• not run per request; the supplied patch was already validated externally
• git apply --check passed against main at c39c433
• git diff --cached --check passed before commit