fix wormhole bench: extend histogram buckets to 30min

[Flotapponnier]

Summary

• Extend wormhole_vaa_latency_milliseconds buckets from 600_000ms cap to 1_800_000ms (30min)
• New buckets: 900_000, 1_200_000, 1_800_000 (15/20/30 min)

Why

Ethereum and Arbitrum VAAs only sign after L1 Casper FFG finality (~12.8min ideal, up to ~19min real). With the previous 600_000ms bucket ceiling, histogram_quantile pinned Eth/Arb p50/p90/p99 all to 600s (the bucket edge), which is not the real latency.

Prior sample distribution: Ethereum 208/219 samples in +Inf, Arbitrum 5/5 in +Inf.

Deploy

1. Merge dev, then cherry-pick to main
2. VPS: rsync harness dir, direct docker build --no-cache, then docker compose up -d --force-recreate wormhole-vaa-latency
3. Wait ~15 min for Ethereum samples to accumulate in new buckets (histogram bucket schema change means old samples are effectively "lost" from a query perspective - they still count but bucket boundaries shifted)

Test plan

• After container recreation, /api/stat/wormhole-vaa-latency returns Ethereum/Arbitrum p50 in ~12-19 min range instead of 600s