chore(infra): add gRPC layer, update protobufs, remove immediate module

- Add gRPC service modules: auth, channel, channel settings, member,
  permission
- Update protobuf definitions and generated code
- Remove immediate/ real-time module (superseded by IM service)
- Update etcd discovery and registration
- Update cache, error, config, and build infrastructure
- Add ADR documentation
- Update OpenAPI spec

docs/adr/002-two-tier-caching.md

@@ -0,0 +1,49 @@
+# ADR-002: 两级缓存架构 / Two-Tier Caching Architecture
+
+## 状态 / Status
+
+**Accepted**
+
+**日期 / Date**: 2024-01-01
+
+## 背景 / Context
+
+平台需要缓存机制来减少数据库负载，提高响应速度。需要在性能和一致性之间取得平衡。
+
+The platform needs a caching mechanism to reduce database load and improve response times. A balance between performance and consistency is required.
+
+## 决策 / Decision
+
+采用 **两级缓存架构**：L1 (内存 LRU-TTL) + L2 (Redis)。
+
+Adopted **two-tier caching**: L1 (in-memory LRU-TTL) + L2 (Redis).
+
+## 考虑的方案 / Considered Options
+
+1. **纯 Redis** — 简单但网络延迟高
+2. **纯内存缓存** — 快但不跨实例共享
+3. **两级缓存** — 兼顾速度和共享
+
+## 后果 / Consequences
+
+### 正面 / Positive
+
+- L1 提供极低延迟 / L1 provides ultra-low latency
+- L2 提供跨实例共享 / L2 provides cross-instance sharing
+- 减少数据库负载 / Reduces database load
+
+### 负面 / Negative
+
+- 缓存一致性复杂 / Cache consistency is complex
+- 内存占用增加 / Increased memory usage
+
+### 风险 / Risks
+
+- 缓存雪崩 / Cache avalanche
+- 缓存穿透 / Cache penetration
+
+## 实现细节 / Implementation Details
+
+- **L1**: `DashMap + Mutex<LruTracker>`, TTL 5 分钟
+- **L2**: Redis via r2d2, TTL 可配置
+- **策略**: L1 miss → L2 miss → 数据库查询