fix(runner): 修复图像结果转存与任务租约失效

修正非语义字段长 Base64 元数据被误判为媒体的问题，并允许按文件签名识别真实媒体后转存 OSS；同时将网关本地存储失败与上游失败分离，避免重复调用和错误降权。

异步任务复用已分配并发名额前立即刷新租约，租约已丢失时重新排队获取，避免排队耗时侵蚀租约后错误调用上游。生产环境关闭 server-main 尚未实现的进度回调，保留 Gateway 任务详情与事件查询。

验证：API 全量 go test、go vet、gofmt、Kubernetes 渲染、迁移安全检查及真实阿里云 OSS 读写/生命周期验收通过。

apps/api/internal/runner/retry_decision_test.go

@@ -495,3 +495,37 @@ func TestResolveCandidateFailureUnmatchedRetryableUsesSameThenNext(t *testing.T)
 		t.Fatalf("unmatched exhausted error should rotate without side effects, got %+v", decision)
 	}
 }
+
+func TestGatewayStorageFailureNeverRetriesProvider(t *testing.T) {
+	err := &clients.ClientError{
+		Code:       "storage_write_failed",
+		Message:    "generated binary result could not be written to object storage",
+		StatusCode: 503,
+		Retryable:  true,
+	}
+	candidate := store.RuntimeModelCandidate{ModelRetryPolicy: map[string]any{
+		"enabled":       true,
+		"allowKeywords": []any{"5xx"},
+	}}
+
+	retry := retryDecisionForCandidate(candidate, err)
+	if retry.Retry || retry.Reason != "result_persistence_failed" {
+		t.Fatalf("Gateway storage failure must not repeat an upstream request: %+v", retry)
+	}
+	failover := failoverDecisionForCandidate(store.RunnerPolicy{
+		Status:         "active",
+		FailoverPolicy: map[string]any{"enabled": true, "allowCategories": []any{"provider_5xx"}},
+	}, candidate, err)
+	if failover.Retry || failover.Reason != "result_persistence_failed" {
+		t.Fatalf("Gateway storage failure must not fail over to another provider: %+v", failover)
+	}
+	decision := resolveCandidateFailure(resolveCandidateFailureInput{
+		RunnerPolicy:     store.RunnerPolicy{Status: "active"},
+		Err:              err,
+		HasNextCandidate: true,
+		Async:            true,
+	})
+	if decision.Route != "stop" || decision.Reason != "result_persistence_failed" || decision.Info.Category != "gateway_storage" {
+		t.Fatalf("Gateway storage failure classification is incorrect: %+v", decision)
+	}
+}