Implement Hyperliquid weighted rate limiter for REST API

[nicolad]

Introduce a weighted, minute-bucket rate limiter for Hyperliquid REST and wire it into Info/Exchange calls. Supports base weights per endpoint, post-response extra debits (for item-scaled endpoints), batch-aware Exchange costs, and bounded retries with full-jitter + Retry-After handling. Surfaces structured RateLimit { scope, weight, retry_after_ms } errors.

Why not reuse NT’s limiter?

• NT’s Quota is uniform per request; Hyperliquid requires weighted charges and post-response debits.
• No batch-size awareness in Quota (Exchange: 1 + floor(n/40)).
• Adapter needs minute bucket + anti-stampede semantics and exchange-specific weight tables.

Implementation highlights

• WeightedLimiter::per_minute(1200): token bucket; acquire(weight), debit_extra(extra), snapshot().
• Weights: info_base_weight, info_extra_weight (largest array, +1/20; candleSnapshot +1/60); exchange_weight.
• Backoff: full-jitter (base=125ms, cap=5s), honor Retry-After; tiny re-acquire to defuse minute rollovers.
• Defaults in HyperliquidHttpClient; optional with_rate_limits() tuning.

[cjdsellers]

Thanks @nicolad, great progress here. We need this because the existing GCRA style rate limiter doesn’t match Hyperliquid’s requirements, and what you’ve implemented here aligns with exactly what's needed.