feat(cag): memory standing-context adapter, warmup semaphore, cache-aware routing

Issue #44: code-register memory adapter with standing-context resource; fix
dispatcher method-level Auth.Capability override (was always using verb-derived
capability, ignoring per-method Auth field).

Issue #45: per-backend bounded in-memory queue (VLLM_MAX_QUEUE_DEPTH=20) +
concurrency semaphore (VLLM_MAX_NUM_SEQS=4); warmup requests (X-Poimen-Warmup:
true) sleep WARMUP_DELAY_MS before queue entry; /metrics endpoint wired.

Issue #46: rendezvous-hash backend selection on X-Poimen-Cache-Key; round-robin
fallback; MarkFailed excludes unhealthy backends 30s; ModifyResponse captures
vllm-num-cached-tokens + emits structured log + increments counter.

Closes #44 Closes #45 Closes #46

internal/server/router.go

@@ -16,6 +16,7 @@ type Router struct {
 	temporalHandler   http.Handler
 	upstreamHandler   http.Handler
 	forgejoWebhook    *webhook.ForgejoHandler
+	metricsHandler    http.Handler
 }
 
 // NewRouter creates a new router with health endpoints.
@@ -33,6 +34,11 @@ func NewRouter(healthChecker *HealthChecker, dispatcher *serviceadapter.Dispatch
 	}
 }
 
+// SetMetricsHandler registers the Prometheus /metrics endpoint handler.
+func (r *Router) SetMetricsHandler(h http.Handler) {
+	r.metricsHandler = h
+}
+
 // ServeHTTP implements http.Handler.
 // Priority order:
 // 1. /healthz and /readyz to health handlers
@@ -48,6 +54,11 @@ func (r *Router) ServeHTTP(w http.ResponseWriter, req *http.Request) {
 	case "/readyz":
 		ReadinessHandler(r.healthChecker)(w, req)
 		return
+	case "/metrics":
+		if r.metricsHandler != nil {
+			r.metricsHandler.ServeHTTP(w, req)
+			return
+		}
 	}
 
 	// X-Service (ServiceAdapter) routing - checked before path-based routing