Build the benchmark harness (sync / execution / RPC, NM + Geth)

[AnkushinDaniil]

Build the reusable, milestone-agnostic benchmark harness that #21 runs at each state size. NM + Geth first (other clients via #26).

Modules

• Sync — drives a fresh client snap-syncing from a milestone source; captures wall time, bandwidth, IOPS, RSS, final DB size (feeds Test snap sync of all EL clients on 5x mainnet bloated state #16/Test snap sync of all EL clients on 7x mainnet bloated state #17/Test snap sync of all EL clients on 10x mainnet bloated state #18)
• Execution — replays a fixed block window; captures block-processing p50/p95/p99, gas/s, CPU%, RSS growth
• RPC latency — concurrent load on eth_getBalance / getStorageAt / call / getProof / getCode; p50/p95/p99
• Output — benchmarks/<milestone>/{client,metric,...}.json|csv; >25% regression flags; Grafana panels
• Repeatable within ±5%

This is the engineering task underneath #21 (which is the runs). Start with the Sync module (it doubles as the Geth-snapshot producer for the cloud deliverable).