Evaluate and implement Session.close()/shutdown()

Old driver: Cluster.shutdown() / Session.shutdown() is used everywhere for deterministic teardown
(test/pylib/scylla_cluster.py:722,731,967,981, test/pylib/driver_utils.py:29).

To evaluate:

• Confirm whether we actually need this, or whether GC-based cleanup is acceptable long-term — but given how load-bearing this pattern is in the old driver's test/fixture code, it's likely genuinely useful (deterministic teardown between tests, and also this would enable properly fixing this bug: https://github.com/scylladb/python-rs-driver/issues/141).
• If we do want it: checked the pinned scylla-rust-driver has no shutdown/close method and no Drop impl on Session either. Cleanup today relies entirely on Rust's normal drop semantics of background tasks/connections. That means this likely can't be solved purely in python-rs-driver. It probably needs an upstream feature in scylla-rust-driver first.
• Open question to resolve: is "just drop it and rely on Drop" actually good enough once wrapped through PyO3 (async runtime + GIL details may make implicit drop-based cleanup less predictable than in pure Rust), or is an explicit close truly required?

Note: In the old driver Cluster implements the plain Python context-manager protocol:enterjust returnsself, and exitjust callsself.shutdown() (cassandra/cluster.py:1907-1911,3283-3287). This is what enables the convenient withblock usage cleanup runs automatically on block exit, even if an exception is raised inside it, with no explicittry/finally` needed at the call site.

Start by reviewing the existing teardown references in test/pylib/scylla_cluster.py and test/pylib/driver_utils.py, then inspect the pinned scylla-rust-driver Session API and its cleanup behavior. Compare this with the old driver's context-manager behavior in cassandra/cluster.py and issue #141. Done means resolving whether implicit drop is sufficient and, if not, identifying the required implementation path and validating deterministic teardown.