feat(acceptance): 增加生产同构媒体压力验收模式

引入动态流量门禁、隔离验收身份与协议级 Gemini/Volces 模拟器，覆盖双站点 API、Worker、PostgreSQL、River、账务、回调和媒体物化链路。

新增 P24/P28/P32 容量阶梯、Worker 强杀恢复、真实小流量 canary、CAS 放量和失败保持 validation 的生产编排；Worker 执行槽、连接池、媒体并发和双站点副本数改为环境配置。

验证：Go 全量测试、真实 PostgreSQL 迁移集成测试、迁移安全检查、OpenAPI 生成、ShellCheck、Kustomize、gofmt 和 git diff --check。

apps/api/internal/httpapi/gemini_base64_stress_integration_test.go

@@ -22,6 +22,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/easyai/easyai-ai-gateway/apps/api/internal/acceptanceworkload"
 	"github.com/easyai/easyai-ai-gateway/apps/api/internal/config"
 	"github.com/easyai/easyai-ai-gateway/apps/api/internal/store"
 	"github.com/jackc/pgx/v5/pgxpool"
@@ -37,13 +38,14 @@ func TestGeminiBase64ImageEditThousandSynchronousRequests(t *testing.T) {
 	}
 
 	const (
-		taskCount          = 1000
 		upstreamConcurrent = 64
 		inputVariants      = 16
 		mediaConcurrent    = 8
 	)
-	imageBytes := stressEnvInt(t, "AI_GATEWAY_GEMINI_STRESS_IMAGE_BYTES", 256<<10)
-	upstreamDelay := time.Duration(stressEnvInt(t, "AI_GATEWAY_GEMINI_STRESS_UPSTREAM_DELAY_MS", 4000)) * time.Millisecond
+	baselineProfile := acceptanceworkload.GeminiBaseline
+	taskCount := baselineProfile.Requests
+	imageBytes := stressEnvInt(t, "AI_GATEWAY_GEMINI_STRESS_IMAGE_BYTES", baselineProfile.InputBytes)
+	upstreamDelay := time.Duration(stressEnvInt(t, "AI_GATEWAY_GEMINI_STRESS_UPSTREAM_DELAY_MS", int(baselineProfile.DelayMin/time.Millisecond))) * time.Millisecond
 	maxHeapGrowth := uint64(stressEnvInt(t, "AI_GATEWAY_GEMINI_STRESS_MAX_HEAP_BYTES", 1<<30))
 	clientConnections := stressEnvInt(t, "AI_GATEWAY_GEMINI_STRESS_CLIENT_CONNECTIONS", 256)
 	testTimeout := time.Duration(stressEnvInt(t, "AI_GATEWAY_GEMINI_STRESS_TIMEOUT_SECONDS", 1200)) * time.Second
@@ -387,7 +389,7 @@ WHERE task.model = $1
 	if tasks != taskCount || succeeded != taskCount || attempts != taskCount || duplicateAttemptTasks != 0 {
 		t.Fatalf("task results tasks=%d succeeded=%d attempts=%d duplicate_attempt_tasks=%d", tasks, succeeded, attempts, duplicateAttemptTasks)
 	}
-	if upstreamCalls.Load() != taskCount || invalidInputs.Load() != 0 {
+	if upstreamCalls.Load() != int64(taskCount) || invalidInputs.Load() != 0 {
 		t.Fatalf("upstream calls=%d invalid_inputs=%d, want %d/0", upstreamCalls.Load(), invalidInputs.Load(), taskCount)
 	}
 	if upstreamPeak.Load() > mediaConcurrent || upstreamPeak.Load() < mediaConcurrent/2 {