Test snap sync of all EL clients on 10x mainnet bloated state

Measure snap sync of the x10 milestone — the headline result. Same metric set + setup as #16, at the x10 state.

Sync pairs — server held constant (NM), syncer varied; + one realism run

• NM → NM — NM sync-cost scaling + snap-synced NM DB size
• NM → Geth — Geth sync-cost scaling; produces the bootable Geth x10 snapshot
• Geth → NM — mainnet-realism run: on mainnet NM syncs mostly from Geth peers, so this checks whether the real-world server mix changes NM's sync cost (high-value at the top milestone)
• (opportunistic) Geth → Geth — server-side effects only

Metrics (per pair)

• Wall time / bandwidth / IOPS / RSS / final DB size

Acceptance

• NM→NM, NM→Geth (and Geth→NM) complete VALID at x10
• Metrics committed to benchmarks/x10/ per pair
• Compared vs x7 (flag >25% regressions)

Depends on #12. Boundary: sync only; exec+RPC in #21.

Start with the setup and metric set referenced by issue #16, then account for the dependency on #12. Run the listed NM→NM, NM→Geth, and Geth→NM snap-sync pairs at the x10 state, recording wall time, bandwidth, IOPS, RSS, and final DB size in benchmarks/x10/; compare each result with x7 and flag regressions above 25%.