feat(worker): 实现集群限流与自适应负载

保留平台模型 RPM、TPM 和并发策略语义，增加 PostgreSQL 集群级租约、饱和候选重选和多平台自动负载，避免突发任务固定等待首个平台。\n\n新增 Worker 实时负载采样、自适应 active/heavy 容量、心跳与管理端指标，并扩展本地 acceptance runner，覆盖三 Worker、同模型三平台 2/4/6 并发和 48 个带图视频突发任务。\n\n验证：go test ./...、go vet ./...、PostgreSQL 跨 Store 集成测试、gofmt、bash -n、ShellCheck 及本地集群 provider-burst 验收通过；48/48 成功，无越限、重复提交、重复计费、重复回调或终态资源泄漏。

apps/api/internal/acceptanceworkload/profiles_test.go

@@ -3,15 +3,18 @@ package acceptanceworkload
 import "testing"
 
 func TestGeminiProfilesAndVideoDistribution(t *testing.T) {
-	for _, profile := range []GeminiProfile{GeminiBaseline, GeminiLarge, GeminiPeak} {
+	for _, profile := range []GeminiProfile{GeminiBaseline, GeminiMultiImage, GeminiLarge, GeminiPeak} {
 		resolved, ok := GeminiProfileByName(profile.Name)
 		if !ok || resolved != profile {
 			t.Fatalf("profile %q resolved to %+v, ok=%v", profile.Name, resolved, ok)
 		}
-		if profile.Requests <= 0 || profile.InputBytes <= 0 || profile.OutputBytes <= 0 ||
+		if profile.Requests <= 0 || profile.InputImages <= 0 || profile.InputBytes < profile.InputImages || profile.OutputBytes < 0 ||
 			profile.DelayMin <= 0 || profile.DelayMax < profile.DelayMin {
 			t.Fatalf("invalid Gemini profile: %+v", profile)
 		}
+		if profile.Name != GeminiMultiImage.Name && profile.OutputBytes == 0 {
+			t.Fatalf("fixed-output Gemini profile has no output size: %+v", profile)
+		}
 	}
 	counts := map[int]int{}
 	for index := 0; index < 100; index++ {