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()`.

Start at Session::open and compare its endpoint loop with reconnect(). Trace authenticate(), including openSession and requestStatementId, to understand where each failure is handled. Done means authentication failures cause the next configured endpoint to be tried, matching reconnect() behavior; no test file is named in the issue.

Written by the indexing model from the issue text.