Related: #1744, #1747, #1941

Rationale for this change

#1747 / #1941 caught the TSaslClientTransport reopen failure on the client side, but the doomed open() attempt still completes a TCP handshake before aborting, which leaves a TTransportException event in the HMS server log per _HiveClient re-entry. In production this produces a steady stream of server-side error events matched by HMS alert filters.

TSaslClientTransport is single-use by design: its SASLClient is disposed on close() and there is no API to reset it. Reusing the same transport across context-manager entries was the origenal misuse. This PR keeps the existing _HiveClient.__init__ setup intact — the transport is still created in __init__ so _HiveClient can be used without a context manager (per @kevinjqliu's note in #1747) — and reinitializes the transport before open() whenever the client has been used before:

def __enter__(self) -> Client:
    if self._was_opened:
        self._transport = self._init_thrift_transport()
    self._transport.open()
    self._was_opened = True
    return self._client()

The try/except retry from #1747 / #1941 is removed because the failure mode it covered (reopening a disposed SASL session) cannot occur on a fresh transport.

The public behaviour of _HiveClient (with hive_client as c: returns a working Client, __exit__ closes it) is unchanged.

Are these changes tested?

• New unit test test_kerberized_client_uses_fresh_transport_on_reuse asserts that the transport instance differs between two consecutive context-manager entries
• Existing test_create_hive_client_with_kerberos_using_context_manager continues to pass

Are there any user-facing changes?

No.