Session::open does not fail over when openSession/requestStatementId fails (only TCP-layer failover)

`Session::open` only fails over at the TCP layer: the loop over `config.endpoints` opens a `Connection`, but `authenticate()` (openSession + requestStatementId) runs **after** the loop. `reconnect()` does the opposite — it retries connect **and** authenticate for every endpoint.

With a 3-node cluster where one node accepts TCP but fails `openSession` (e.g. overloaded, misconfigured, or wedged by the missing socket timeout), roughly one `open()` in three fails while 2 of 3 nodes are healthy.

Fix: move the `authenticate()` step inside the endpoint loop, mirroring `reconnect()`.

從 Session::open 開始，將其端點迴圈與 reconnect() 進行比較。追蹤 authenticate()，包括 openSession 和 requestStatementId，以了解每個失敗在哪裡得到處理。完成的標準是：驗證失敗會導致嘗試下一個已設定的端點，與 reconnect() 的行為一致；issue 中未指定任何測試檔案。

由索引模型根據 Issue 內容生成。