xchain fairness v3: bucket + guard alignment across 4 e2e harnesses

[Flotapponnier]

Deep-review finding: LayerZero/Hyperlane/Axelar were dropping messages >30min as noise + capping histograms at 30min, while CCIP kept 2h drop + 90min buckets. Meant p90/p99 comparisons in the meta-bench favored the 3 protocols artificially (their tails were trimmed, CCIP's wasn't).

Fixes

1. Unify maxLatencyMs across 4 e2e harnesses

• LayerZero: 30min → 2h
• Hyperlane: 30min → 2h
• Axelar: 30min → 2h
• CCIP: already 2h
  Same drop threshold for all four.

2. Unify histogram buckets (16 buckets, 5s to 90min)
Same bucket layout as CCIP for all four. Preserves the tail for honest p99.

3. Fairness methodology update
Corrected the "which protocols wait for finality" description:

• Was: "CCIP waits, LZ/Hyperlane/Axelar sign pre-finality"
• Now: "CCIP AND Axelar both wait on Ethereum-source; LZ + Hyperlane sign pre-finality"
  Axelar's ~96-confirmation wait on ETH-source is documented in both the meta-bench and the dedicated Axelar bench.

Impact

Data reset: rebuild+recreate 3 containers on VPS drops accumulated 1-2h of history in old bucket layout. Fine — worker will refill within an hour.

Deploy

• All specs validate (94 specs)
• All 3 harnesses build
• VPS: rebuild LZ + Hyperlane + Axelar containers, rebuild worker