fix(saas-relay): eliminate DATABASE_ERROR by removing DB queries from critical path

Root cause: each relay request executes 13-17 serial DB queries, exhausting
the 50-connection pool under concurrency. When pool is exhausted, sqlx returns
PoolTimedOut which maps to 500 DATABASE_ERROR.

Fixes:
1. log_operation → dispatch_log_operation (async Worker dispatch, non-blocking)
2. record_usage → tokio::spawn (3 DB queries moved off critical path)
3. DB pool: max_connections 50→100 (env-configurable), acquire_timeout 5s→8s

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

crates/zclaw-saas/src/db.rs

@@ -8,10 +8,22 @@ const SCHEMA_VERSION: i32 = 11;
 
 /// 初始化数据库
 pub async fn init_db(database_url: &str) -> SaasResult<PgPool> {
+    // 连接池大小可通过环境变量配置，默认 100（relay 请求每次 10+ 串行查询，50 偏紧）
+    let max_connections: u32 = std::env::var("ZCLAW_DB_MAX_CONNECTIONS")
+        .ok()
+        .and_then(|v| v.parse().ok())
+        .unwrap_or(100);
+    let min_connections: u32 = std::env::var("ZCLAW_DB_MIN_CONNECTIONS")
+        .ok()
+        .and_then(|v| v.parse().ok())
+        .unwrap_or(5);
+
+    tracing::info!("Database pool: max={}, min={}", max_connections, min_connections);
+
     let pool = PgPoolOptions::new()
-        .max_connections(50)
-        .min_connections(3)
-        .acquire_timeout(std::time::Duration::from_secs(5))
+        .max_connections(max_connections)
+        .min_connections(min_connections)
+        .acquire_timeout(std::time::Duration::from_secs(8))
         .idle_timeout(std::time::Duration::from_secs(180))
         .max_lifetime(std::time::Duration::from_secs(900))
         .connect(database_url)