Fix: Connection pool should not close connections assigned to queued requests - #1114

Summary

Fixes #1110

Problem

There is a race condition in the connection pool where _assign_requests_to_connections() can close a connection that has already been assigned to a queued request.

The race condition:

1. Thread A assigns an idle connection to a queued PoolRequest via �ssign_to_connection(). The connection remains IDLE because handle_request() hasn't been called yet.
2. Thread B calls _assign_requests_to_connections() and sees the connection as still IDLE. It closes the connection as surplus keep-alive or to make room for a new connection.
3. Thread A wakes up and calls connection.handle_request() on the now-closed connection:
   • If the request hasn't started → raises ConnectionNotAvailable → request re-queued
   • If the request has started → the socket is closed under a request waiting for its response → hangs until read timeout

Fix

In the surplus idle connection cleanup loop within _assign_requests_to_connections(), skip connections that have been assigned to any request in the pool.

Added _is_connection_assigned() helper method to both ConnectionPool (sync) and AsyncConnectionPool (async) that checks whether a connection is referenced by any request in the pool's request list.

Changes

• httpcore/_sync/connection_pool.py: Added _is_connection_assigned() method; modified surplus idle cleanup to exclude assigned connections
• httpcore/_async/connection_pool.py: Same fix for async variant
• ests/_sync/test_connection_pool_issue1110.py: Regression tests (4 tests)
• ests/_async/test_connection_pool_issue1110.py: Async regression tests (4 tests)

Testing

All 12 new tests pass. All existing sync connection pool tests pass (4 failures are pre-existing: 3 missing h2 module, 1 keepalive_expiry timing).