Track deterministic device-performance regressions

[asteinh]

Summary

• compare committed Cortex-M and ESP32-S3 silicon summaries with the PR or push base revision
• report every latency, all-in memory, plan/model-size, and firmware-size delta while rejecting only material regressions
• reject successful cells or tracked metrics disappearing, with focused mutation coverage
• keep the gate hardware-free and avoid timing shared CI runners

Validation

• clean-clone unit suite: 35 passed, 1 optional sibling-check skip
• tracked JSON and provenance validation against develop
• shell syntax and diff checks

No benchmark result, core pin, dependency, runtime code, schema, or target footprint changes.