fix: data-api report corrections from live bench audit

[Flotapponnier]

What changed

All prose numbers replaced with live bench API values (Aug 5 2026). Structural and factual errors from the internal audit fixed.

Number corrections

• Jupiter quote: 96.6% → 97.1%
• KyberSwap quote: 92.9% → 87.8%
• Mobula quote: 76.4% → 71.3%
• OpenSea NFT: 93.2% → 93.8%
• Alchemy NFT: 73.7% → 74.3%
• XRPScan wallet labels: now correctly leads StellarExpert (80.0% vs 79.9%)
• Moralis wallet labels: was missing entirely, added at 44.4%
• All remaining wallet label figures corrected to live values
• Global Mobula head lag: 707ms → 677ms

Factual fixes

• "CoinGecko more than twice CoinPaprika" → "1.5x" (461/307)
• "OLI marginally ahead of Blockscout" → inverted (Blockscout 55.6% > OLI 50.2%)
• "Jupiter highest absolute figure in cohort" → Jupiter and Moralis tied at 97.1%
• NFT gap: qualified to "between a cohort leader and a major general-purpose incumbent"
• "six categories" → "seven categories" throughout

Solana section rewrite
The block-time causation story was factually wrong: BNB (0.45s blocks post-Fermi Jan 2026), Base (0.2s Flashblocks), and Robinhood Chain (0.1s) all have faster or equal blocks to Solana (0.4s), yet their head lags are 8-10x higher. Section rewritten as "Head Lag Reflects Integration Depth, Not Block Speed" with the actual anti-correlation shown in the table. RH Chain disclaimer added (launched July 1, 2026 — 34 days at time of report).