fix(saas): reduce DB connection pool pressure in relay path

1. key_pool: merge 3 serial UPDATE queries into 2 (cumulative stats +
   last_used_at combined into single UPDATE)
2. service: reduce SSE spawn sleep from 3s to 500ms and add 5s timeout
   on DB operations to prevent connection hoarding

crates/zclaw-saas/src/relay/key_pool.rs

@@ -137,6 +137,7 @@ pub async fn select_best_key(db: &PgPool, provider_id: &str, enc_key: &[u8; 32])
 }
 
 /// 记录 Key 使用量（滑动窗口）
+/// 合并为 2 次查询：1 次 upsert 滑动窗口 + 1 次更新 provider_keys 累计统计（含 last_used_at）
 pub async fn record_key_usage(
     db: &PgPool,
     key_id: &str,
@@ -144,6 +145,7 @@ pub async fn record_key_usage(
 ) -> SaasResult<()> {
     let current_minute = chrono::Utc::now().format("%Y-%m-%dT%H:%M").to_string();
 
+    // 1. Upsert sliding window
     sqlx::query(
         "INSERT INTO key_usage_window (key_id, window_minute, request_count, token_count)
          VALUES ($1, $2, 1, $3)
@@ -154,20 +156,18 @@ pub async fn record_key_usage(
     .bind(key_id).bind(&current_minute).bind(tokens.unwrap_or(0))
     .execute(db).await?;
 
-    // 更新 Key 的累计统计
+    // 2. Update cumulative stats + last_used_at in one query
     sqlx::query(
-        "UPDATE provider_keys SET total_requests = total_requests + 1, total_tokens = total_tokens + COALESCE($1, 0), updated_at = $2
-         WHERE id = $3"
+        "UPDATE provider_keys
+         SET total_requests = total_requests + 1,
+             total_tokens = total_tokens + COALESCE($1, 0),
+             last_used_at = NOW(),
+             updated_at = NOW()
+         WHERE id = $2"
     )
-    .bind(tokens).bind(&chrono::Utc::now().to_rfc3339()).bind(key_id)
+    .bind(tokens).bind(key_id)
     .execute(db).await?;
 
-    // 更新最后使用时间 (LRU 排序依据)
-    sqlx::query("UPDATE provider_keys SET last_used_at = NOW() WHERE id = $1")
-        .bind(key_id)
-        .execute(db)
-        .await?;
-
     Ok(())
 }