perf(queue): 限制大媒体任务内存并消除准入惊群

将同步与异步非文本任务的准入唤醒改为按任务和全局队首推进，批量续租等待者，避免大量请求同时争抢 advisory lock 和数据库连接。

对 Base64 请求解析、素材解码、上游媒体执行和结果物化增加分层并发限制，复用请求素材并释放重复 Gemini wire 数据；生产 API 默认入口解析并发 16，媒体物化并发 8。

新增 1000 个同步 Gemini 图像编辑请求的模拟上游压力验收，10 MiB 输入和输出下全部成功，Heap 峰值增长约 2.58 GiB，并验证共享上传哈希、单次 attempt 与本地零落盘。

验证：go test ./... -count=1；go vet ./...；gofmt -l 无输出；kubectl kustomize deploy/kubernetes/production；10 MiB Gemini Base64 千任务压力测试通过。

apps/api/internal/store/rate_limit_policy_test.go

@@ -1,6 +1,10 @@
 package store
 
-import "testing"
+import (
+	"errors"
+	"testing"
+	"time"
+)
 
 func TestNormalizeAndValidateQueuePolicy(t *testing.T) {
 	t.Run("defaults maximum wait and preserves extensions", func(t *testing.T) {
@@ -227,3 +231,64 @@ func TestAsyncWorkerCapacityRespectsUserGroupCeiling(t *testing.T) {
 		})
 	}
 }
+
+func TestAdmissionDeadlineSeparatesExecutionSlotWaitFromPolicyQueueTimeout(t *testing.T) {
+	now := time.Now()
+	infrastructureDeadline := admissionDeadline([]AdmissionScope{{
+		ConcurrentLimit: 10,
+	}})
+	if wait := infrastructureDeadline.Sub(now); wait < 23*time.Hour || wait > executionSlotWaitMax+time.Second {
+		t.Fatalf("execution-slot deadline wait=%s, want about %s", wait, executionSlotWaitMax)
+	}
+
+	defaultPolicyDeadline := admissionDeadline([]AdmissionScope{{
+		ConcurrentLimit: 10,
+		QueueLimit:      20,
+	}})
+	if wait := defaultPolicyDeadline.Sub(now); wait < 599*time.Second || wait > 601*time.Second {
+		t.Fatalf("default policy deadline wait=%s, want about 600s", wait)
+	}
+
+	explicitPolicyDeadline := admissionDeadline([]AdmissionScope{{
+		ConcurrentLimit: 10,
+		QueueLimit:      20,
+		MaxWaitSeconds:  90,
+	}})
+	if wait := explicitPolicyDeadline.Sub(now); wait < 89*time.Second || wait > 91*time.Second {
+		t.Fatalf("explicit policy deadline wait=%s, want about 90s", wait)
+	}
+}
+
+func TestQueuedAdmissionCountsFIFOWaitersAsVirtualCapacity(t *testing.T) {
+	scope := AdmissionScope{
+		ScopeType:       "platform_model",
+		ScopeKey:        "model-1",
+		ConcurrentLimit: 1,
+		Amount:          1,
+		QueueLimit:      2,
+	}
+	for waiting := 0; waiting <= 2; waiting++ {
+		if err := validateQueuedAdmissionCapacity([]admissionScopeState{{
+			Scope:   scope,
+			Waiting: waiting,
+		}}); err != nil {
+			t.Fatalf("waiting=%d rejected within concurrent=1 queue=2: %v", waiting, err)
+		}
+	}
+	err := validateQueuedAdmissionCapacity([]admissionScopeState{{
+		Scope:   scope,
+		Waiting: 3,
+	}})
+	var limitErr *RateLimitExceededError
+	if !errors.As(err, &limitErr) || limitErr.Reason != "queue_full" {
+		t.Fatalf("fourth task error=%v, want queue_full", err)
+	}
+
+	scope.QueueLimit = 0
+	if err := validateQueuedAdmissionCapacity([]admissionScopeState{{
+		Scope:   scope,
+		Waiting: 1,
+	}}); !errors.As(err, &limitErr) || limitErr.Metric != "concurrent" {
+		t.Fatalf("queue-disabled second task error=%v, want concurrent rate limit", err)
+	}
+}