feat: add truthful benchmark phase telemetry

[shaggitza]

Summary

• publish schema-v4 runner manifests with exact preparation and cold-build phase boundaries
• mark warm, incremental, RSS, and cache-size telemetry explicitly not measured
• authenticate producer/evaluator status, timing keys, phase order, aggregates, and unresolved normalization fail closed
• preserve schema-v3 compatibility and prevent writes into frozen benchmark artifacts/results

Validation

• 62 focused runner/evaluator tests passed
• adversarial complete-manifest, cleanup-failure, phase-order, extra-timing, and frozen-path cases passed
• Ruff, strict mypy, clean-index and frozen-artifact checks passed
• four independent review rounds; final verdict PASS

Partial progress on #283; real warm/incremental backend reuse, RSS/cache measurement, and remaining release-attestation gates stay open.