feat(peg): OHLC per-minute gauges + worst-case headline metric

[Flotapponnier]

Summary

Fix A of the methodology improvements proposed by a Coinpaprika data
team reviewer. Switches stablecoin-peg from per-minute median to
per-minute MAX deviation, so sub-minute depeg wicks (e.g. USDC at $0.87
for 5 seconds during SVB) stop disappearing into the median.

What changed

Harness (Go):

• New gauges in metrics.go: peg_deviation_worst_bps, peg_minute_max_bps, _min_bps, _open_bps, _close_bps
• aggregator.gocloseMinuteLocked() now computes OHLC across every venue sample within the 60s bucket and emits all five gauges alongside the existing median

YAML (benchmarks/stablecoin-peg.yml):

• All 3 provider queries (USDC, USDT, DAI) point at peg_deviation_worst_bps
• Methodology section explains the OHLC switch + the rationale
• Comment block updated to list all new metrics

What did NOT change

• peg_deviation_bps (legacy median) stays live for backward compat
• Histogram + cross-venue gap + depeg flag unchanged
• No outlier rule changes here (separate PR — Fix B)
• No site code changes (Next.js side resolves metrics dynamically from YAML)

Test plan

• Harness binary builds clean (go build ./... from harnesses/stablecoin-peg/)
• Deploy to Railway, scrape /metrics, confirm all 5 new gauges emit non-zero values per stable
• Verify peg_deviation_worst_bps >= peg_deviation_bps for every minute (max >= median)
• Bench page on staging shows updated leaderboard with the new headline
• Prometheus query quantile_over_time(0.99, peg_deviation_worst_bps[24h]) returns valid numbers within ~24h of deploy