[All] Graceful close on arrow streams

[elenagaljak-db]

What changes are proposed in this pull request?

• Adds stream_paused_max_wait_time_ms config option to ArrowStreamConfigurationOptions across all SDKs (Rust, Go, Python, TypeScript, Java)
• Implements client-side graceful close state machine: when the server signals it will close the stream, the SDK enters a paused state, stops sending new batches but drains in-flight ack responses before triggering recovery
• Graceful close recoveries don't count against the recovery_retries budget, since they're intentional server-side restarts (e.g. rolling deploys)

How is this tested?

• 5 new integration tests in rust/tests/src/arrow_tests.rs covering: full server duration wait, immediate recovery, client-capped wait, early exit when all batches acked, ingest accepted during grace period