audit round 2: gram methodology, buyback percent + success

[Flotapponnier]

Three findings from the continued audit, each verified before changing:

1. Gram finality 0.4s is CORRECT (empirically measured 0.42s/block; masterchain is BFT-final at production since the accelerated upgrade) but the methodology text described a 3-seqno lookback the measurement does not use. Text now describes the real measurement; the external reference claiming ~2.7s/block is pre-upgrade.
2. buyback-audit rendered raw fractions with a percent sign (Hyperliquid 0.94%): ratios now scaled x100 in queries, page reads 94%.
3. Sky success showed 0% while its data was live and fresh: the formula multiplied any scrape error rate by 300, and coingecko throttling (327 lifetime errors, price still cached fine) zeroed it. Success is now scrape freshness (last success within 2h), matching what the column claims to mean.