Problem

The existing near_limit counter fires at a fixed percentage controlled by the global NEAR_LIMIT_RATIO env var (default 80%). There is no way to:

1. Set a different threshold per-service/descriptor (e.g. alert at 60% for a critical service).
2. Build a utilization ratio in Prometheus because the configured requests_per_unit value is never exported as a metric — the denominator simply doesn't exist.

Solution

Add a rate_limit gauge to RateLimitStats that holds the configured requests_per_unit for each descriptor. The gauge is populated once at config load time inside NewRateLimit(), so it reflects the current config and is updated on hot-reload.

With this gauge, operators can construct arbitrary utilization alerts in Prometheus without touching NEAR_LIMIT_RATIO:

# Alert when any descriptor exceeds 60% utilization
ratelimit_service_rate_limit_total_hits
  / ratelimit_service_rate_limit_limit > 0.6

# Per-domain alert at a custom threshold
ratelimit_service_rate_limit_total_hits{domain="payments"}
  / ratelimit_service_rate_limit_limit{domain="payments"} > 0.9

Unlimited descriptors (requests_per_unit: 0) emit 0, which produces +Inf/NaN in ratio queries and does not trigger threshold alerts.

Changes

Test plan

• go test ./test/stats/... — new TestNewStatsCreatesRateLimitGauge passes
• go test ./test/config/... ./test/limiter/... ./test/redis/... ./test/memcached/... ./test/service/... — all existing tests pass
• go build ./... — compiles cleanly