win 95814974de feat(rpm): add token bucket smoothing for RPM rate limiting
- New RPMTokenBucketService: per-account continuous-refill token buckets
  (rate = rpm/60 tokens/sec, capacity = rpm). No new dependencies.
- GatewayService.AcquireRPMToken() delegates to the bucket service.
- Gateway handler inserts RPM token wait BEFORE wrapReleaseOnDone in both
  Gemini and Anthropic dispatch paths; timeout returns 429 and releases slot.
- Config: gateway.rpm_smoothing.enabled (default false) + max_wait_ms (default 5000).
- 7 unit tests covering: immediate acquire, zero RPM, timeout, wait+refill,
  context cancel, account isolation, bucket reset on RPM change.
2026-04-29 01:22:54 +08:00
..
2026-03-09 01:18:49 +08:00