fix: transport use-after-close crashes append session

[quettabit]

resolves #92

an example for a use-after-close scenario that can cause this crash is calling transport's close twice.

[quettabit]

@greptileai review pls

[greptile-apps]

Greptile Summary

The PR prevents failed or closing transports from remaining eligible for pooled reuse and consolidates HTTP/2 writes behind one drain-aware flush path.

• Adds send-side health and byte-count state to each connection.
• Marks connections unavailable when socket writes or drains fail.
• Uses forced drains for lifecycle-sensitive writes and threshold-based drains for data and acknowledgment traffic.
• Updates the stream-ID allocation test for the consolidated flush method.

Confidence Score: 5/5

The PR appears safe to merge.

The changed flush path now catches socket write and drain failures, marks the connection send-dead, and excludes it from subsequent pool checkout, so no blocking failure remains.

Important Files Changed

Filename	Overview
src/s2_sdk/_client.py	Consolidates flushing and marks transports send-dead after socket failures, addressing the previously reported connection-reuse path.
tests/test_client.py	Updates the existing stream-allocation test to mock the replacement flush method.

_{Reviews (3): Last reviewed commit: "initial commit" | Re-trigger Greptile}

[quettabit]

@greptileai addressed your comments. pls review again.