perf(worker): 按实例内存上限扩展异步容量

将策略推导出的集群目标与单 Worker 内存安全上限分离，避免失活实例的全部容量转移到单个 2 GiB Pod。

生产环境单实例上限和媒体物化并发设为 24；两实例正常时总容量由 16 提升至 48，后续可通过增加 Worker 节点继续扩展。新增 0093 迁移保存实例容量上限，并兼容滚动发布中的旧实例。

风险：更高媒体并发会增加 Worker 和节点内存压力，保留单实例 24 的硬边界并由持续监控观察 RSS、OOM、队列与节点余量。

验证：go test ./... -count=1；go vet ./...；真实 PostgreSQL 分配与故障转移测试；迁移安全检查；kubectl kustomize；gofmt -l；git diff --check。

scripts/migration-safety-reviewed.json

@@ -8,6 +8,13 @@
         "non-null column addition"
       ],
       "reason": "The migration replaces two existing constraints and one index with expanded definitions. Production preflight confirmed six schema-v1 identity revisions, no OIDC sessions or revocation watermarks, and small affected relations; the non-null columns use constant defaults."
+    },
+    "apps/api/migrations/0093_worker_instance_capacity_limit.sql": {
+      "sha256": "4b6a39ef90e7032555739ccf0568bf7ae0bce23ef98c3bee9c2c8e3bfa13291e",
+      "allowedViolations": [
+        "non-null column addition"
+      ],
+      "reason": "The table stores ephemeral Worker heartbeats and currently has only two production rows. PostgreSQL adds the integer column with a constant default, then the migration backfills each live row from its existing allocation so a rolling deployment cannot assign unsafe failover capacity."
     }
   }
 }